<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Gavindu achintha</title>
    <description>The latest articles on DEV Community by Gavindu achintha (@gavinduachintha).</description>
    <link>https://dev.to/gavinduachintha</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F1636782%2F85203a0a-1ecc-4d37-a7cc-2ec273c13fa0.png</url>
      <title>DEV Community: Gavindu achintha</title>
      <link>https://dev.to/gavinduachintha</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/gavinduachintha"/>
    <language>en</language>
    <item>
      <title>Haunted Terminal: a local spirit that can change your prompt, but not your shell</title>
      <dc:creator>Gavindu achintha</dc:creator>
      <pubDate>Sat, 31 Oct 2026 07:31:36 +0000</pubDate>
      <link>https://dev.to/gavinduachintha/haunted-terminal-a-local-spirit-that-can-change-your-prompt-but-not-your-shell-jhp</link>
      <guid>https://dev.to/gavinduachintha/haunted-terminal-a-local-spirit-that-can-change-your-prompt-but-not-your-shell-jhp</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week4-2026-10-26"&gt;Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;Haunted Terminal is a Python REPL with a ghost in it. You type ordinary commands and they run on your machine. Between commands, a spirit comments, and the prompt itself changes mood and personality.&lt;/p&gt;

&lt;p&gt;The week asked for something spooky, or something that honors the past. I wanted both. The Poltergeist rattles the pipes when a command fails. The Archivist keeps a candle for finished work, in the spirit of Día de los Muertos: failed commands, deleted files, and commits are names worth remembering. The Keeper watches the prompt the way a lighthouse keeper watches weather.&lt;/p&gt;

&lt;p&gt;Each spirit is a TOML file. Switching is &lt;code&gt;/spirit archivist&lt;/code&gt;. A new ghost is a new file, not a code change.&lt;/p&gt;

&lt;p&gt;The model is Gemma, served by Ollama on &lt;code&gt;127.0.0.1&lt;/code&gt;. The default tag is &lt;code&gt;gemma3:1b&lt;/code&gt;. If that tag is not pulled, &lt;code&gt;--model&lt;/code&gt; points at whichever Gemma you already have. If Ollama is asleep, or a reply takes longer than the timeout, the panel says the spirit is whispering from memory and prints a canned line. The prompt comes back either way.&lt;/p&gt;

&lt;p&gt;Rituals start with &lt;code&gt;/&lt;/code&gt; and never reach the operating system:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;/manifest&lt;/code&gt; writes one ghost file under &lt;code&gt;.haunt/manifestations/&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;/banish&lt;/code&gt; deletes only that folder&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;/epitaph&lt;/code&gt; drafts a commit message and runs &lt;code&gt;git commit&lt;/code&gt; only after you type &lt;code&gt;y&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;/quiet&lt;/code&gt; and &lt;code&gt;/loud&lt;/code&gt; turn the commentary off and on&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;There is no website. The product is the terminal. &lt;code&gt;haunt --demo&lt;/code&gt; records two files.&lt;/p&gt;

&lt;p&gt;This first transcript uses the canned voice, so it replays with the network unplugged. The panels say "from memory" on purpose.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;poltergeist (rattling):~/haunt-demo λ ls no-such-candle
ls: cannot access 'no-such-candle': No such file or directory
┌─ Poltergeist whispers from memory ─┐
│ I knocked twice on the return key. The exit code knocked back harder. │
└─┘

/spirit archivist
Archivist takes the prompt.

/manifest
A file appeared: .haunt/manifestations/ofrenda.txt

/epitaph
Commit with this epitaph?

Remember what this change tried to keep

Set among the other finished work.
The archivist speaks the subject aloud once, then lets the log hold it.

Type y to commit:
y
Committed: Remember what this change tried to keep

/quiet
git status
On branch master
nothing to commit, working tree clean
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;/quiet&lt;/code&gt; is the tell. &lt;code&gt;git status&lt;/code&gt; still ran. The Archivist did not speak over it.&lt;/p&gt;

&lt;p&gt;The second file is one live call. This machine already had &lt;code&gt;gemma3:latest&lt;/code&gt; on disk, so that is the tag in the recording. The Archivist was given the command &lt;code&gt;ls no-such-candle&lt;/code&gt; and exit code 2, and answered:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;The flame flickers, acknowledging the lost name. Another echo joins the quiet vigil.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Command output stayed on the terminal. The model saw the command text, the exit code, and the directory name.&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/Gavinduachintha/Spooky" rel="noopener noreferrer"&gt;https://github.com/Gavinduachintha/Spooky&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The piece that makes the project work is the local call, and the wall around it. &lt;code&gt;voice.py&lt;/code&gt; is the only module that knows the Ollama URL, and it accepts &lt;code&gt;127.0.0.1&lt;/code&gt;, &lt;code&gt;localhost&lt;/code&gt;, or &lt;code&gt;::1&lt;/code&gt;. Before a prompt is built, anything that is not the command, the exit code, a short git summary, or a few recent commands is dropped. That includes stdout and stderr.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;shrink_context&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;context&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nb"&gt;dict&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;-&amp;gt;&lt;/span&gt; &lt;span class="nb"&gt;dict&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="sh"&gt;"""&lt;/span&gt;&lt;span class="s"&gt;Keep the prompt small. Unknown keys, including command output, are dropped.&lt;/span&gt;&lt;span class="sh"&gt;"""&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Every request sets &lt;code&gt;num_ctx&lt;/code&gt; to 4096, caps the reply (about 70 tokens for a whisper), and keeps the model loaded with &lt;code&gt;keep_alive&lt;/code&gt;. Ghost files have to survive a path check before they are written. &lt;code&gt;/epitaph&lt;/code&gt; passes the subject and body as &lt;code&gt;git commit -m&lt;/code&gt; arguments. It does not push, amend, or force.&lt;/p&gt;

&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;Open-weight Gemma on Ollama is the core, not a caption. Four tasks share one client:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;a whisper, temperature 0.9, short&lt;/li&gt;
&lt;li&gt;a one- or two-word mood for the prompt, cached so redrawing the prompt does not hit the model&lt;/li&gt;
&lt;li&gt;a ghost file, as JSON &lt;code&gt;{filename, body}&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;a commit message, as JSON &lt;code&gt;{subject, body}&lt;/code&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The HTTP call is &lt;code&gt;urllib&lt;/code&gt; against &lt;code&gt;POST /api/chat&lt;/code&gt;. &lt;a href="https://dev.to/damiano_builds"&gt;Damiano&lt;/a&gt; wrote up why a small local tool often does not need an SDK for this, and also how slow a small Gemma can be on CPU: on the order of tens of seconds for a long answer. A terminal cannot wait that long, so the timeout defaults to 8 seconds and the canned line is the success path, not an error page. Streaming still fills the panel when the model is fast enough.&lt;/p&gt;

&lt;p&gt;Structured tasks use Ollama's JSON schema on &lt;code&gt;/api/chat&lt;/code&gt;, which is the shape &lt;a href="https://dev.to/amareswer/ollama-api-a-practical-guide-with-examples-4di9"&gt;Amareswer's Ollama API guide&lt;/a&gt; walks through: chat, options, and a &lt;code&gt;format&lt;/code&gt; object when the reply has to be data.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://dev.to/nazar-boyko/running-local-llms-with-ollama-for-private-development-4924"&gt;Nazar Boyko&lt;/a&gt; is the reason &lt;code&gt;num_ctx&lt;/code&gt; is set on every request. Ollama's default window is 2048 tokens, and it clips the rest with no error. These prompts stay short so the spirit is reacting to the command you just typed.&lt;/p&gt;

&lt;p&gt;The REPL is &lt;code&gt;prompt_toolkit&lt;/code&gt;. Whispers are Rich panels. &lt;code&gt;cd&lt;/code&gt; and &lt;code&gt;pwd&lt;/code&gt; are remembered by the session, because every other command starts a fresh shell. Twenty-six tests run without a network: a fake HTTP handler stands in for Ollama, path escapes have to stay inside the sandbox, and a declined epitaph must not create a commit.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;A spirit that comments on your shell is a bad fit for a hosted model. The text you type includes paths, hostnames, repo names, and the occasional secret you did not mean to type. That stream belongs on the machine that already has the terminal.&lt;/p&gt;

&lt;p&gt;A closed API would also be the wrong shape for the rest of it. Commentary is frequent, so a per-token bill is a tax on using your own prompt. The weights stay useful with the network off after the one-time &lt;code&gt;ollama pull&lt;/code&gt;. The tag is a flag, so the same spirits can speak through &lt;code&gt;gemma3:1b&lt;/code&gt;, &lt;code&gt;gemma3:270m&lt;/code&gt;, or &lt;code&gt;gemma3:latest&lt;/code&gt; without a rewrite. And a local model will stay in character when the command is &lt;code&gt;kill&lt;/code&gt; or a failing &lt;code&gt;ls&lt;/code&gt;, which is the whole joke.&lt;/p&gt;

&lt;p&gt;The open piece is what makes the haunting honest. The ghost can hear the command. It cannot leave the room.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Best Use of Gemma.&lt;/strong&gt; Whispers, the prompt's mood, ghost files, and commit messages are Gemma 3 through Ollama on loopback. No hosted model and no API key.&lt;/p&gt;

</description>
      <category>hf26challenge</category>
      <category>devchallenge</category>
      <category>python</category>
      <category>ai</category>
    </item>
    <item>
      <title>MailTriage: a local inbox desk a freelancer would pay for</title>
      <dc:creator>Gavindu achintha</dc:creator>
      <pubDate>Sat, 17 Oct 2026 06:23:42 +0000</pubDate>
      <link>https://dev.to/gavinduachintha/mailtriage-a-local-inbox-desk-a-freelancer-would-pay-for-55dp</link>
      <guid>https://dev.to/gavinduachintha/mailtriage-a-local-inbox-desk-a-freelancer-would-pay-for-55dp</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week2-2026-10-12"&gt;Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;MailTriage&lt;/strong&gt; is a local triage desk for freelancers. You connect Gmail, the app sorts client mail by what actually needs you, and a local model drafts a reply in your voice. You edit it. Nothing is sent until you save it to Gmail Drafts yourself.&lt;/p&gt;

&lt;p&gt;Categories: urgent, waiting on a client, invoice, new lead, low priority. Each message also gets a priority score, a one-line summary, and a suggested action.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Who would pay:&lt;/strong&gt; a freelancer whose inbox is invoices, “just circling back”, real deadlines, and new leads — and who will not paste client names and rates into a hosted chatbot.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The problem:&lt;/strong&gt; cloud copilots want the whole thread. MailTriage keeps inference on the laptop (Ollama + Gemma 3). Gmail tokens stay in a signed cookie. Postgres only stores writing-style examples and the last triage per message, not the inbox.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Gmail  --read only--&amp;gt;  MailTriage  --local Gemma--&amp;gt;  draft
                            |
                       Postgres
                  (style + last triage)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;



&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fduvjsf1ylc0wm5jfh5lo.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fduvjsf1ylc0wm5jfh5lo.png" alt=" " width="800" height="379"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2h6gopu2vfz37m3nioqa.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2h6gopu2vfz37m3nioqa.png" alt=" " width="695" height="705"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxaspt4kc1jcuv000f4v0.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxaspt4kc1jcuv000f4v0.png" alt=" " width="668" height="208"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffv50h58ipo7ote4uxv57.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffv50h58ipo7ote4uxv57.png" alt=" " width="648" height="289"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fa3y2js4kldjfhl35f7th.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fa3y2js4kldjfhl35f7th.png" alt=" " width="661" height="525"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;




&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/Gavinduachintha" rel="noopener noreferrer"&gt;
        Gavinduachintha
      &lt;/a&gt; / &lt;a href="https://github.com/Gavinduachintha/mailtriage" rel="noopener noreferrer"&gt;
        mailtriage
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;p&gt;This is a &lt;a href="https://nextjs.org" rel="nofollow noopener noreferrer"&gt;Next.js&lt;/a&gt; project bootstrapped with &lt;a href="https://nextjs.org/docs/app/api-reference/cli/create-next-app" rel="nofollow noopener noreferrer"&gt;&lt;code&gt;create-next-app&lt;/code&gt;&lt;/a&gt;.&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Getting Started&lt;/h2&gt;
&lt;/div&gt;

&lt;p&gt;First, run the development server:&lt;/p&gt;

&lt;div class="highlight highlight-source-shell notranslate position-relative overflow-auto js-code-highlight"&gt;
&lt;pre&gt;npm run dev
&lt;span class="pl-c"&gt;&lt;span class="pl-c"&gt;#&lt;/span&gt; or&lt;/span&gt;
yarn dev
&lt;span class="pl-c"&gt;&lt;span class="pl-c"&gt;#&lt;/span&gt; or&lt;/span&gt;
pnpm dev
&lt;span class="pl-c"&gt;&lt;span class="pl-c"&gt;#&lt;/span&gt; or&lt;/span&gt;
bun dev&lt;/pre&gt;

&lt;/div&gt;
&lt;p&gt;Open &lt;a href="http://localhost:3000" rel="nofollow noopener noreferrer"&gt;http://localhost:3000&lt;/a&gt; with your browser to see the result.&lt;/p&gt;
&lt;p&gt;You can start editing the page by modifying &lt;code&gt;app/page.tsx&lt;/code&gt;. The page auto-updates as you edit the file.&lt;/p&gt;
&lt;p&gt;This project uses &lt;a href="https://nextjs.org/docs/app/building-your-application/optimizing/fonts" rel="nofollow noopener noreferrer"&gt;&lt;code&gt;next/font&lt;/code&gt;&lt;/a&gt; to automatically optimize and load &lt;a href="https://vercel.com/font" rel="nofollow noopener noreferrer"&gt;Geist&lt;/a&gt;, a new font family for Vercel.&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Learn More&lt;/h2&gt;
&lt;/div&gt;

&lt;p&gt;To learn more about Next.js, take a look at the following resources:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;a href="https://nextjs.org/docs" rel="nofollow noopener noreferrer"&gt;Next.js Documentation&lt;/a&gt; - learn about Next.js features and API.&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://nextjs.org/learn" rel="nofollow noopener noreferrer"&gt;Learn Next.js&lt;/a&gt; - an interactive Next.js tutorial.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;You can check out &lt;a href="https://github.com/vercel/next.js" rel="noopener noreferrer"&gt;the Next.js GitHub repository&lt;/a&gt; - your feedback and contributions are welcome!&lt;/p&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Deploy on Vercel&lt;/h2&gt;
&lt;/div&gt;

&lt;p&gt;The easiest way to deploy your Next.js app is to use the &lt;a href="https://vercel.com/new?utm_medium=default-template&amp;amp;filter=next.js&amp;amp;utm_source=create-next-app&amp;amp;utm_campaign=create-next-app-readme" rel="nofollow noopener noreferrer"&gt;Vercel Platform&lt;/a&gt; from the creators of Next.js.&lt;/p&gt;

&lt;p&gt;Check out our &lt;a href="https://nextjs.org/docs/app/building-your-application/deploying" rel="nofollow noopener noreferrer"&gt;Next.js deployment documentation&lt;/a&gt; for more…&lt;/p&gt;&lt;/div&gt;


&lt;/div&gt;
&lt;br&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/Gavinduachintha/mailtriage" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;br&gt;
&lt;/div&gt;
&lt;br&gt;


&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;Open-source AI is the core, not a plugin:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Local inference:&lt;/strong&gt; Ollama on &lt;code&gt;127.0.0.1&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Open-weight model:&lt;/strong&gt; Gemma 3 (&lt;code&gt;OLLAMA_MODEL=gemma3&lt;/code&gt;)&lt;/li&gt;
&lt;li&gt;The model does the whole job: category, priority, summary, suggested action, and the draft. Style examples you save (past replies) are loaded from Postgres and sent with the prompt so the next draft sounds like you.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The rest of the stack is a desk around that call:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Next.js 16&lt;/strong&gt; — inbox list, retractable sidebar, draft panel, settings&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gmail API&lt;/strong&gt; — read-only inbox, save to Drafts (never auto-send)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;PostgreSQL + &lt;code&gt;pg&lt;/code&gt;&lt;/strong&gt; — no ORM. Tables are &lt;code&gt;mailtriage_users&lt;/code&gt;, &lt;code&gt;mailtriage_style_examples&lt;/code&gt;, &lt;code&gt;mailtriage_analyses&lt;/code&gt; so they do not collide with an existing &lt;code&gt;users&lt;/code&gt; table. Queries are plain &lt;code&gt;SELECT&lt;/code&gt; / &lt;code&gt;INSERT&lt;/code&gt; / &lt;code&gt;UPDATE&lt;/code&gt; / &lt;code&gt;DELETE&lt;/code&gt;. Saving an analysis is an update, then an insert if nothing matched.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The Next.js pool is one &lt;code&gt;pg.Pool&lt;/code&gt; on &lt;code&gt;globalThis&lt;/code&gt; so hot reload does not leak connections. If &lt;code&gt;DATABASE_URL&lt;/code&gt; changes, the pool is rebuilt.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;A freelancer will not pay for a tool that uploads their clients to a hosted model. Rates, NDAs, and “can you invoice this week” do not belong in someone else’s cache.&lt;/p&gt;

&lt;p&gt;A closed mail API would have been easier to demo (paste a key, ship a URL) and worse to sell. Open-weight Gemma on Ollama is what makes the side hustle honest:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Privacy&lt;/strong&gt; — the thread is inferred on the same machine that already has the inbox&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cost&lt;/strong&gt; — no per-token bill for “sorry for the delay,” so the product can be a one-time or cheap local tool&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Control&lt;/strong&gt; — swap the model with &lt;code&gt;OLLAMA_MODEL&lt;/code&gt; instead of rebuilding around one vendor’s price and terms&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The open pieces are why someone would pay &lt;em&gt;this&lt;/em&gt; instead of pasting the email into ChatGPT.&lt;/p&gt;

&lt;h2&gt;
  
  
  My Agent Session
&lt;/h2&gt;


&lt;div class="ltag-agent-session"&gt;
  &lt;div class="agent-session-header"&gt;
    
    &lt;span class="agent-session-tool-icon-badge" title="GitHub Copilot"&gt;
&lt;/span&gt;
    &lt;span class="agent-session-title"&gt;Building MailTriage: local Gemma triage, Postgres, Week 2 post&lt;/span&gt;
  &lt;/div&gt;

  &lt;div class="agent-session-scroll"&gt;

      &lt;div class="agent-session-message agent-session-user"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-user"&gt;
          You
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Analyze the UI. It looks like the usual LLM dashboard. Make it professional for freelancers, keep it consistent, no fancy gradients.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-assistant"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-assistant"&gt;
          Agent
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Replaced the indigo/serif LLM look with warm stone neutrals, a single muted teal accent, IBM Plex Sans/Mono wired correctly, and a calmer inbox/sidebar/draft panel. No gradients.&lt;/p&gt;

                &lt;/div&gt;
              &lt;div class="agent-session-tool-call"&gt;
                
                  &lt;span class="agent-session-chevron"&gt;▸&lt;/span&gt;
                  &lt;span class="tool-name tool-name-searchreplace"&gt;search_replace&lt;/span&gt;
                    &lt;span class="tool-input-preview"&gt;app/globals.css palette + font-sans&lt;/span&gt;
                
                &lt;div class="agent-session-tool-detail"&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Input&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;app/globals.css palette + font-sans&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Output&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;Stone surfaces, teal primary, Geist mismatch fixed to Plex&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                &lt;/div&gt;
              &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-user"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-user"&gt;
          You
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Add a settings option and make the sidebar retractable.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-assistant"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-assistant"&gt;
          Agent
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Desktop sidebar collapses to an icon rail (saved on the device). Settings dialog covers theme, writing style, local model status, and Gmail disconnect.&lt;/p&gt;

                &lt;/div&gt;
              &lt;div class="agent-session-tool-call"&gt;
                
                  &lt;span class="agent-session-chevron"&gt;▸&lt;/span&gt;
                  &lt;span class="tool-name tool-name-write"&gt;write&lt;/span&gt;
                    &lt;span class="tool-input-preview"&gt;components/sidebar.tsx + settings-dialog.tsx&lt;/span&gt;
                
                &lt;div class="agent-session-tool-detail"&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Input&lt;/div&gt;
                      &lt;pre class="agent-session-pre highlight tsx"&gt;&lt;code&gt;&lt;span class="nx"&gt;components&lt;/span&gt;&lt;span class="o"&gt;/&lt;/span&gt;&lt;span class="nx"&gt;sidebar&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;tsx&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="nx"&gt;settings&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nx"&gt;dialog&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;tsx&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Output&lt;/div&gt;
                      &lt;pre class="agent-session-pre highlight tsx"&gt;&lt;code&gt;&lt;span class="nx"&gt;Collapse&lt;/span&gt; &lt;span class="nx"&gt;toggle&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="nx"&gt;Settings&lt;/span&gt; &lt;span class="nx"&gt;under&lt;/span&gt; &lt;span class="nx"&gt;Workspace&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                &lt;/div&gt;
              &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-user"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-user"&gt;
          You
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Add an avatar in the header. The application name is MailTriage.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-assistant"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-assistant"&gt;
          Agent
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Header shows initials from the Gmail address and a menu (settings, disconnect). Renamed InboxGuard to MailTriage in the UI, metadata, landing copy, and model prompt.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-user"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-user"&gt;
          You
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Add a database. Using Postgres. Keep it simple, no ORM.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-assistant"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-assistant"&gt;
          Agent
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Added pg Pool + raw SQL. Tables: users, style examples, analyses. Gmail tokens stay in the cookie. Inbox still comes from Gmail.&lt;/p&gt;

                &lt;/div&gt;
              &lt;div class="agent-session-tool-call"&gt;
                
                  &lt;span class="agent-session-chevron"&gt;▸&lt;/span&gt;
                  &lt;span class="tool-name tool-name-write"&gt;write&lt;/span&gt;
                    &lt;span class="tool-input-preview"&gt;lib/db/pool.ts, lib/db/queries.ts, db/schema.sql&lt;/span&gt;
                
                &lt;div class="agent-session-tool-detail"&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Input&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;lib/db/pool.ts, lib/db/queries.ts, db/schema.sql&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Output&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;Parameterized queries, no Drizzle&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                &lt;/div&gt;
              &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-user"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-user"&gt;
          You
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Keep the db query simple. I need the full query including creating the tables.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-assistant"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-assistant"&gt;
          Agent
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Flattened to SELECT/INSERT/UPDATE/DELETE. Save path is UPDATE then INSERT. Gave the full CREATE TABLE script.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-user"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-user"&gt;
          You
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;CREATE TABLE users skipped (already exists). style_examples failed: column email referenced in foreign key does not exist.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-assistant"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-assistant"&gt;
          Agent
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;They were on postgres.public. An existing users table had no email column. Renamed tables to mailtriage_users, mailtriage_style_examples, mailtriage_analyses.&lt;/p&gt;

                &lt;/div&gt;
              &lt;div class="agent-session-tool-call"&gt;
                
                  &lt;span class="agent-session-chevron"&gt;▸&lt;/span&gt;
                  &lt;span class="tool-name tool-name-write"&gt;write&lt;/span&gt;
                    &lt;span class="tool-input-preview"&gt;prefixed table names&lt;/span&gt;
                
                &lt;div class="agent-session-tool-detail"&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Input&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;prefixed table names&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Output&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;FK now references mailtriage_users(email)&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                &lt;/div&gt;
              &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-user"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-user"&gt;
          You
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;password authentication failed for user postgres&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-assistant"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-assistant"&gt;
          Agent
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;A direct pg client connected. Next.js was still using a pool created with the old DATABASE_URL. Pool now rebuilds when the URL changes and prefers &lt;span class="agent-session-redacted"&gt;[REDACTED]&lt;/span&gt;. Restart the dev server after env changes.&lt;/p&gt;

                &lt;/div&gt;
              &lt;div class="agent-session-tool-call"&gt;
                
                  &lt;span class="agent-session-chevron"&gt;▸&lt;/span&gt;
                  &lt;span class="tool-name tool-name-write"&gt;write&lt;/span&gt;
                    &lt;span class="tool-input-preview"&gt;lib/db/pool.ts recreate on URL change&lt;/span&gt;
                
                &lt;div class="agent-session-tool-detail"&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Input&lt;/div&gt;
                      &lt;pre class="agent-session-pre highlight typescript"&gt;&lt;code&gt;&lt;span class="nx"&gt;lib&lt;/span&gt;&lt;span class="o"&gt;/&lt;/span&gt;&lt;span class="nx"&gt;db&lt;/span&gt;&lt;span class="o"&gt;/&lt;/span&gt;&lt;span class="nx"&gt;pool&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;ts&lt;/span&gt; &lt;span class="nx"&gt;recreate&lt;/span&gt; &lt;span class="nx"&gt;on&lt;/span&gt; &lt;span class="nx"&gt;URL&lt;/span&gt; &lt;span class="nx"&gt;change&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Output&lt;/div&gt;
                      &lt;pre class="agent-session-pre highlight typescript"&gt;&lt;code&gt;&lt;span class="nx"&gt;Stale&lt;/span&gt; &lt;span class="nx"&gt;pool&lt;/span&gt; &lt;span class="nx"&gt;no&lt;/span&gt; &lt;span class="nx"&gt;longer&lt;/span&gt; &lt;span class="nx"&gt;keeps&lt;/span&gt; &lt;span class="nx"&gt;a&lt;/span&gt; &lt;span class="nx"&gt;bad&lt;/span&gt; &lt;span class="nx"&gt;password&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                &lt;/div&gt;
              &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-user"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-user"&gt;
          You
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Create the markdown file. Add aesthetics. This is for Hacktoberfest Week 2. Put it on DEV too. Here is the official submission template.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-assistant"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-assistant"&gt;
          Agent
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Wrote README.md for the Side Hustle pitch. Drafted a DEV post on the linked account with the Week 2 template, repo embed, and Best Use of Gemma. Left it unpublished for review.&lt;/p&gt;

                &lt;/div&gt;
              &lt;div class="agent-session-tool-call"&gt;
                
                  &lt;span class="agent-session-chevron"&gt;▸&lt;/span&gt;
                  &lt;span class="tool-name tool-name-createarticle"&gt;create_article&lt;/span&gt;
                    &lt;span class="tool-input-preview"&gt;title MailTriage Week 2 draft, published false&lt;/span&gt;
                
                &lt;div class="agent-session-tool-detail"&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Input&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;title MailTriage Week 2 draft, published false&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Output&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;Draft saved; edit_url handed to user&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                &lt;/div&gt;
              &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;
  &lt;/div&gt;

  &lt;div class="agent-session-footer"&gt;
    &lt;span class="agent-session-meta"&gt;
        16 of 16 messages
    &lt;/span&gt;
  &lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;Build log from this session: UI pass, retractable sidebar and settings, MailTriage rename, Postgres without an ORM, the &lt;code&gt;users&lt;/code&gt; table collision, and the stale connection-pool password bug.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Best Use of Gemma&lt;/strong&gt; — triage and drafts run on Gemma 3 through Ollama. No hosted LLM.&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>[Boost]</title>
      <dc:creator>Gavindu achintha</dc:creator>
      <pubDate>Mon, 12 Oct 2026 13:20:03 +0000</pubDate>
      <link>https://dev.to/gavinduachintha/-43ea</link>
      <guid>https://dev.to/gavinduachintha/-43ea</guid>
      <description>&lt;div class="ltag__link--embedded"&gt;
  &lt;div class="crayons-story "&gt;
  &lt;a href="https://dev.to/gavinduachintha/i-built-a-fully-offline-nature-field-guide-with-gemma-3-1i43" class="crayons-story__hidden-navigation-link"&gt;I built a fully offline nature field guide with Gemma 3&lt;/a&gt;


  &lt;div class="crayons-story__body crayons-story__body-full_post"&gt;
      &lt;a href="https://dev.to/gavinduachintha/i-built-a-fully-offline-nature-field-guide-with-gemma-3-1i43" class="crayons-article__context-note crayons-article__context-note__feed"&gt;&lt;p&gt;Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass Submission 🌿&lt;/p&gt;

&lt;/a&gt;
    &lt;div class="crayons-story__top"&gt;
      &lt;div class="crayons-story__meta"&gt;
        &lt;div class="crayons-story__author-pic"&gt;

          &lt;a href="/gavinduachintha" class="crayons-avatar  crayons-avatar--l  "&gt;
            &lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F1636782%2F85203a0a-1ecc-4d37-a7cc-2ec273c13fa0.png" alt="gavinduachintha profile" class="crayons-avatar__image"&gt;
          &lt;/a&gt;
        &lt;/div&gt;
        &lt;div&gt;
          &lt;div&gt;
            &lt;a href="/gavinduachintha" class="crayons-story__secondary fw-medium m:hidden"&gt;
              Gavindu achintha
            &lt;/a&gt;
            &lt;div class="profile-preview-card relative mb-4 s:mb-0 fw-medium hidden m:inline-block"&gt;
              
                Gavindu achintha
                
                
              
              &lt;div id="story-author-preview-content-4831817" class="profile-preview-card__content crayons-dropdown branded-7 p-4 pt-0"&gt;
                &lt;div class="gap-4 grid"&gt;
                  &lt;div class="-mt-4"&gt;
                    &lt;a href="/gavinduachintha" class="flex"&gt;
                      &lt;span class="crayons-avatar crayons-avatar--xl mr-2 shrink-0"&gt;
                        &lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F1636782%2F85203a0a-1ecc-4d37-a7cc-2ec273c13fa0.png" class="crayons-avatar__image" alt=""&gt;
                      &lt;/span&gt;
                      &lt;span class="crayons-link crayons-subtitle-2 mt-5"&gt;Gavindu achintha&lt;/span&gt;
                    &lt;/a&gt;
                  &lt;/div&gt;
                  &lt;div class="print-hidden"&gt;
                    
                      Follow
                    
                  &lt;/div&gt;
                  &lt;div class="author-preview-metadata-container"&gt;&lt;/div&gt;
                &lt;/div&gt;
              &lt;/div&gt;
            &lt;/div&gt;

          &lt;/div&gt;
          &lt;a href="https://dev.to/gavinduachintha/i-built-a-fully-offline-nature-field-guide-with-gemma-3-1i43" class="crayons-story__tertiary fs-xs"&gt;&lt;time&gt;Oct 11&lt;/time&gt;&lt;span class="time-ago-indicator-initial-placeholder"&gt;&lt;/span&gt;&lt;/a&gt;
        &lt;/div&gt;
      &lt;/div&gt;

    &lt;/div&gt;

    &lt;div class="crayons-story__indention"&gt;
      &lt;h2 class="crayons-story__title crayons-story__title-full_post"&gt;
        &lt;a href="https://dev.to/gavinduachintha/i-built-a-fully-offline-nature-field-guide-with-gemma-3-1i43" id="article-link-4831817"&gt;
          I built a fully offline nature field guide with Gemma 3
        &lt;/a&gt;
      &lt;/h2&gt;
        &lt;div class="crayons-story__tags"&gt;
            &lt;a class="crayons-tag  crayons-tag--monochrome " href="/t/devchallenge"&gt;&lt;span class="crayons-tag__prefix"&gt;#&lt;/span&gt;devchallenge&lt;/a&gt;
            &lt;a class="crayons-tag  crayons-tag--monochrome " href="/t/hf26challenge"&gt;&lt;span class="crayons-tag__prefix"&gt;#&lt;/span&gt;hf26challenge&lt;/a&gt;
        &lt;/div&gt;
      &lt;div class="crayons-story__bottom"&gt;
        &lt;div class="crayons-story__details"&gt;
          &lt;a href="https://dev.to/gavinduachintha/i-built-a-fully-offline-nature-field-guide-with-gemma-3-1i43" class="crayons-btn crayons-btn--s crayons-btn--ghost crayons-btn--icon-left"&gt;
            &lt;div class="multiple_reactions_aggregate"&gt;
              &lt;span class="multiple_reactions_icons_container"&gt;
                  &lt;span class="crayons_icon_container"&gt;
                    &lt;img src="https://assets.dev.to/assets/fire-f60e7a582391810302117f987b22a8ef04a2fe0df7e3258a5f49332df1cec71e.svg" width="18" height="18"&gt;
                  &lt;/span&gt;
              &lt;/span&gt;
              &lt;span class="aggregate_reactions_counter"&gt;2&lt;span class="hidden s:inline"&gt;&amp;nbsp;reactions&lt;/span&gt;&lt;/span&gt;
            &lt;/div&gt;
          &lt;/a&gt;
            &lt;a href="https://dev.to/gavinduachintha/i-built-a-fully-offline-nature-field-guide-with-gemma-3-1i43#comments" class="crayons-btn crayons-btn--s crayons-btn--ghost crayons-btn--icon-left flex items-center"&gt;
              

              &lt;span class="hidden s:inline"&gt;Add&amp;nbsp;Comment&lt;/span&gt;
            &lt;/a&gt;
        &lt;/div&gt;
        &lt;div class="crayons-story__save"&gt;
          &lt;small class="crayons-story__tertiary fs-xs mr-2"&gt;
            3 min read
          &lt;/small&gt;
        &lt;/div&gt;
      &lt;/div&gt;
    &lt;/div&gt;
  &lt;/div&gt;
&lt;/div&gt;

&lt;/div&gt;


</description>
      <category>ai</category>
      <category>gemma</category>
      <category>opensource</category>
    </item>
    <item>
      <title>I built a fully offline nature field guide with Gemma 3</title>
      <dc:creator>Gavindu achintha</dc:creator>
      <pubDate>Sun, 11 Oct 2026 08:02:51 +0000</pubDate>
      <link>https://dev.to/gavinduachintha/i-built-a-fully-offline-nature-field-guide-with-gemma-3-1i43</link>
      <guid>https://dev.to/gavinduachintha/i-built-a-fully-offline-nature-field-guide-with-gemma-3-1i43</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week1-2026-10-05"&gt;Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;Trailside is an on-device nature field guide. Point your camera at a plant, bird, insect, or fungus — or upload a photo you already took — and get field notes in seconds. No signal, no cloud, no account. Everything runs locally with a small vision model, so it works on the trail, in the woods, or anywhere the cell bars disappear.&lt;/p&gt;

&lt;p&gt;You can use it from your laptop or phone: open the app in the browser, connect your device camera to snap something in the moment, or drop in a local image from your gallery. Optionally add a quick note of what you already know, and Trailside returns structured field notes. Save the ones that matter to a personal LogBook so the memory stays with the place, not the feed.&lt;/p&gt;

&lt;p&gt;It is built for people who want to look closer at the living world instead of scrolling past it: hikers, gardeners, kids on weekend walks, anyone who has ever wondered “what is that?” and wished for a quiet answer in their pocket.&lt;/p&gt;

&lt;p&gt;The goal is simple: get people off the screen and back into the world — with just enough help to notice more of what is already around them.&lt;/p&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://youtu.be/t1g7IHkAPrk" rel="noopener noreferrer"&gt;Video Demo&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftwfmmcngkds26zcix275.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftwfmmcngkds26zcix275.png" alt="Trailside home screen" width="800" height="384"&gt;&lt;/a&gt;&lt;br&gt;
&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frigu5u1d3p49h7tya63i.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frigu5u1d3p49h7tya63i.png" alt="Field notes in progress" width="799" height="381"&gt;&lt;/a&gt;&lt;br&gt;
&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fhiam5ojb8trhdblutt95.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fhiam5ojb8trhdblutt95.png" alt="LogBook view" width="800" height="721"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/Gavinduachintha/Trailside" rel="noopener noreferrer"&gt;https://github.com/Gavinduachintha/Trailside&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;Trailside is built around &lt;strong&gt;Ollama&lt;/strong&gt; running the open-weight &lt;strong&gt;Gemma 3&lt;/strong&gt; vision model (&lt;code&gt;gemma3:latest&lt;/code&gt;) entirely on-device. There is no cloud API and no remote inference — the model lives on the user’s machine, so identification works offline on the trail.&lt;/p&gt;

&lt;p&gt;The app is a &lt;strong&gt;Streamlit&lt;/strong&gt; front end that handles camera capture or file upload, lightly processes the image with &lt;strong&gt;Pillow&lt;/strong&gt; (EXIF orientation, resize, JPEG encode), then streams the photo and an optional user hint into Ollama’s chat API. A system prompt and a small prompt builder shape the model’s reply into readable field notes. Responses are streamed token-by-token into a journal-style panel so the experience feels immediate.&lt;/p&gt;

&lt;p&gt;Saved sightings (image + notes + optional hint) are stored in a local database via a thin &lt;code&gt;config.db&lt;/code&gt; layer, and the LogBook view lets users browse, open, and delete entries. The whole stack — UI, local model, and storage — stays on the device so the tool remains useful where signal is weak or absent.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;Open innovation is what makes Trailside possible as a real field tool instead of another phone feature that dies when the bars disappear.&lt;/p&gt;

&lt;p&gt;Because the model is open-weight and runs locally through Ollama, every photo stays on the device. There is no upload, no account, no usage quota, and no dependency on a remote API that could change price, rate limits, or terms overnight. That privacy and reliability matter when you are identifying something in the woods — not because the data is sensitive in a corporate sense, but because the tool should work where you actually need it: offline, on a trail, with whatever hardware you already have.&lt;/p&gt;

&lt;p&gt;A closed vision API would have forced a constant network connection, sent images off-device, and tied the project to someone else’s pricing and availability. Open models and local inference remove that gate. They let a small, single-purpose app treat the model as a quiet companion on the path rather than a service you rent by the request.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Best Use of Gemma&lt;/strong&gt; — Trailside is built around Gemma 3 running locally through Ollama. The model does the full multimodal identification: it receives the photo (and optional user hint), streams the field notes, and never leaves the device. No other AI or partner runtime is involved.&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>My friends refuse to write Dockerfiles, so I built an AI that does it for them. You're welcome, lazy people. 🐳</title>
      <dc:creator>Gavindu achintha</dc:creator>
      <pubDate>Sun, 04 Oct 2026 07:01:37 +0000</pubDate>
      <link>https://dev.to/gavinduachintha/dockforge-ai-powered-dockerfile-generator-3hm1</link>
      <guid>https://dev.to/gavinduachintha/dockforge-ai-powered-dockerfile-generator-3hm1</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-weekend-2026-10-01"&gt;Hacktoberfest Weekend Challenge: Build for a Friend&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h1&gt;
  
  
  What I Built
&lt;/h1&gt;

&lt;p&gt;I built &lt;strong&gt;DockForge&lt;/strong&gt;, an AI-powered Docker engineer that turns a software repository into a working Docker setup.&lt;/p&gt;

&lt;p&gt;I built it for a friend who often works on projects that &lt;em&gt;run perfectly on their machine&lt;/em&gt; but become painful to containerize.&lt;/p&gt;

&lt;p&gt;Instead of manually figuring out the runtime, dependencies, package manager, ports, build commands, system packages, and startup configuration, DockForge analyzes the repository and generates a Docker setup tailored to the project.&lt;/p&gt;

&lt;p&gt;The goal isn't just to generate a Dockerfile that &lt;em&gt;looks correct&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;DockForge is designed around a &lt;strong&gt;generate → build → verify → fix&lt;/strong&gt; workflow.&lt;/p&gt;

&lt;p&gt;It can:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Analyze the repository structure and identify the technology stack&lt;/li&gt;
&lt;li&gt;Detect dependencies and package managers&lt;/li&gt;
&lt;li&gt;Determine likely build and start commands&lt;/li&gt;
&lt;li&gt;Generate a Dockerfile&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;In other words:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Repository → Analyze → AI → Dockerfile → Build&lt;/strong&gt;&lt;/p&gt;




&lt;h1&gt;
  
  
  Demo
&lt;/h1&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2g9gl1xoat5298wfg5yt.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2g9gl1xoat5298wfg5yt.png" alt=" " width="800" height="381"&gt;&lt;/a&gt;&lt;/p&gt;


&lt;div class="crayons-card c-embed text-styles text-styles--secondary"&gt;
    &lt;div class="c-embed__content"&gt;
      &lt;div class="c-embed__body flex items-center justify-between"&gt;
        &lt;a href="https://dockforge-chi.vercel.app/" rel="noopener noreferrer" class="c-link fw-bold flex items-center"&gt;
          &lt;span class="mr-2"&gt;dockforge-chi.vercel.app&lt;/span&gt;
          

        &lt;/a&gt;
      &lt;/div&gt;
    &lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;The demo shows DockForge taking an existing repository and automatically producing a Docker configuration instead of requiring the developer to manually write one from scratch.&lt;/p&gt;




&lt;h1&gt;
  
  
  Code
&lt;/h1&gt;


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/Gavinduachintha" rel="noopener noreferrer"&gt;
        Gavinduachintha
      &lt;/a&gt; / &lt;a href="https://github.com/Gavinduachintha/Dockforge" rel="noopener noreferrer"&gt;
        Dockforge
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      Ai powered docker file generating companion
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;div&gt;
&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;🐳 DockForge&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/33a0fbb7385ea6fca5fb85fd74473ce9a85fe57be2ade4a38de628ac774d00ce/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f446f636b466f7267652d76312e302e302d626c61636b3f7374796c653d666f722d7468652d6261646765266c6f676f3d646f636b6572"&gt;&lt;img src="https://camo.githubusercontent.com/33a0fbb7385ea6fca5fb85fd74473ce9a85fe57be2ade4a38de628ac774d00ce/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f446f636b466f7267652d76312e302e302d626c61636b3f7374796c653d666f722d7468652d6261646765266c6f676f3d646f636b6572" alt="DockForge Banner"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/54b851e11ebc5897a8066fe02fb773b2682a5e4003d35b5ad5027bf55283a0d2/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4e6578742e6a732d31352d626c61636b3f7374796c653d666f722d7468652d6261646765266c6f676f3d6e6578742e6a73"&gt;&lt;img src="https://camo.githubusercontent.com/54b851e11ebc5897a8066fe02fb773b2682a5e4003d35b5ad5027bf55283a0d2/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4e6578742e6a732d31352d626c61636b3f7374796c653d666f722d7468652d6261646765266c6f676f3d6e6578742e6a73" alt="Next.js"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/e68ca76fd7bc972f242e145e867b9ed75166bf9a3d5a8242b675e8e07cd91bd1/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f41492d506f77657265642d4646384145323f7374796c653d666f722d7468652d6261646765266c6f676f3d6f70656e6169"&gt;&lt;img src="https://camo.githubusercontent.com/e68ca76fd7bc972f242e145e867b9ed75166bf9a3d5a8242b675e8e07cd91bd1/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f41492d506f77657265642d4646384145323f7374796c653d666f722d7468652d6261646765266c6f676f3d6f70656e6169" alt="AI Powered"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/aeab11b25054de160642d99db104f65cbf5306958288f6ac44b67f197ae1c4bc/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4c6963656e73652d4d49542d737563636573733f7374796c653d666f722d7468652d6261646765"&gt;&lt;img src="https://camo.githubusercontent.com/aeab11b25054de160642d99db104f65cbf5306958288f6ac44b67f197ae1c4bc/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4c6963656e73652d4d49542d737563636573733f7374796c653d666f722d7468652d6261646765" alt="License"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/50c57b723324fa657a266e0955883955609d6fa967cecb8597cc8e72b672914c/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4861636b746f626572666573742d323032342d6f72616e67653f7374796c653d666f722d7468652d6261646765"&gt;&lt;img src="https://camo.githubusercontent.com/50c57b723324fa657a266e0955883955609d6fa967cecb8597cc8e72b672914c/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4861636b746f626572666573742d323032342d6f72616e67653f7374796c653d666f722d7468652d6261646765" alt="Hacktoberfest"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;AI-Powered Dockerfile Generator for Modern Applications&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;Analyze any GitHub repository and get an optimized, production-ready Dockerfile in seconds.&lt;/p&gt;
&lt;p&gt;&lt;a href="https://github.com/Gavinduachintha/Dockforge#-quick-start" rel="noopener noreferrer"&gt;🚀 Quick Start&lt;/a&gt; • &lt;a href="https://github.com/Gavinduachintha/Dockforge/SETUP_GUIDE.md" rel="noopener noreferrer"&gt;📖 Setup Guide&lt;/a&gt; • &lt;a href="https://github.com/Gavinduachintha/Dockforge#-demo" rel="noopener noreferrer"&gt;🎮 Demo&lt;/a&gt; • &lt;a href="https://github.com/Gavinduachintha/Dockforge#-features" rel="noopener noreferrer"&gt;🛠️ Features&lt;/a&gt; • &lt;a href="https://github.com/Gavinduachintha/Dockforge#-contributing" rel="noopener noreferrer"&gt;🤝 Contributing&lt;/a&gt;&lt;/p&gt;
&lt;/div&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;🎯 What is DockForge?&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;DockForge is an intelligent tool that analyzes your GitHub repository and automatically generates production-ready Dockerfiles tailored to your project's specific needs. No more copy-pasting generic Dockerfiles or spending hours optimizing Docker configurations!&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h3 class="heading-element"&gt;Why DockForge?&lt;/h3&gt;
&lt;/div&gt;
&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td width="25%"&gt;
&lt;div class="markdown-heading"&gt;
&lt;h3 class="heading-element"&gt;⚡ Fast&lt;/h3&gt;

&lt;/div&gt;
&lt;p&gt;Generate Dockerfiles in seconds, not hours&lt;/p&gt;
&lt;/td&gt;
&lt;td width="25%"&gt;
&lt;div class="markdown-heading"&gt;
&lt;h3 class="heading-element"&gt;🧠 Smart&lt;/h3&gt;

&lt;/div&gt;
&lt;p&gt;AI-powered analysis and optimization&lt;/p&gt;
&lt;/td&gt;
&lt;td width="25%"&gt;
&lt;div class="markdown-heading"&gt;
&lt;h3 class="heading-element"&gt;🎯 Accurate&lt;/h3&gt;

&lt;/div&gt;
&lt;p&gt;Framework and dependency detection&lt;/p&gt;
&lt;/td&gt;
&lt;td width="25%"&gt;
&lt;div class="markdown-heading"&gt;
&lt;h3 class="heading-element"&gt;✨ Modern&lt;/h3&gt;

&lt;/div&gt;
&lt;p&gt;Best practices and security built-in&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;🚀 Quick Start&lt;/h2&gt;

&lt;/div&gt;
&lt;p&gt;Get DockForge running in &lt;strong&gt;5 minutes&lt;/strong&gt;:&lt;/p&gt;
&lt;div class="highlight highlight-source-shell notranslate position-relative overflow-auto js-code-highlight"&gt;
&lt;pre&gt;&lt;span class="pl-c"&gt;&lt;span class="pl-c"&gt;#&lt;/span&gt; 1️⃣ Clone the repository&lt;/span&gt;
git clone https://github.com/yourusername/dockforge.git
&lt;span class="pl-c1"&gt;cd&lt;/span&gt; dockforge
&lt;span class="pl-c"&gt;&lt;span class="pl-c"&gt;#&lt;/span&gt; 2️⃣ Install dependencies&lt;/span&gt;
npm install

&lt;span class="pl-c"&gt;&lt;span class="pl-c"&gt;#&lt;/span&gt; 3️⃣ Set up environment variables&lt;/span&gt;
cp .env.example .env
&lt;span class="pl-c"&gt;&lt;span class="pl-c"&gt;#&lt;/span&gt; Add your OpenRouter API key to .env&lt;/span&gt;

&lt;span class="pl-c"&gt;&lt;span class="pl-c"&gt;#&lt;/span&gt; 4️⃣ Start the development server&lt;/span&gt;&lt;/pre&gt;…
&lt;/div&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/Gavinduachintha/Dockforge" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;
&lt;br&gt;
DockForge is built as an open-source project, so the implementation can be inspected, modified, and extended by other developers.


&lt;h1&gt;
  
  
  How I Built It
&lt;/h1&gt;

&lt;p&gt;I built the whole application with Next.js and used the OpenRouter API to connect it with an LLM.&lt;/p&gt;

&lt;p&gt;For the AI side, I'm using &lt;strong&gt;Qwen3.8-27B&lt;/strong&gt; whick is free. DockForge takes the project information and asks the model to figure out what kind of Docker setup the project needs and generate the Dockerfile.&lt;/p&gt;

&lt;p&gt;I wanted to keep the idea simple: instead of spending time figuring out how to Dockerize a project manually, you can give it to DockForge and let the AI handle the first draft for you.&lt;/p&gt;

&lt;p&gt;OpenRouter also made it really easy to experiment with different models without having to change the whole application.&lt;/p&gt;

&lt;p&gt;The process is roughly:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Repository
    ↓
Repository Analysis
    ↓
Project Detection
    ↓
AI Planning
    ↓
Dockerfile Generation
    ↓
Docker Build
    ↓
Build Errors / Verification
    ↓
AI Fixes
    ↓
Working Container
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The AI is responsible for reasoning about the project and generating the configuration, while Docker provides the real-world feedback.&lt;/p&gt;

&lt;p&gt;This distinction is important.&lt;/p&gt;

&lt;p&gt;A language model can generate a Dockerfile that &lt;em&gt;sounds right&lt;/em&gt; while still being completely broken. By actually attempting to build the image, DockForge can use Docker's output as feedback and iterate on the configuration.&lt;/p&gt;

&lt;p&gt;For the AI layer, I used &lt;strong&gt;open-weight models through OpenRouter&lt;/strong&gt;, allowing the underlying model to be replaced without redesigning the application around a single proprietary AI provider.&lt;/p&gt;

&lt;p&gt;This also means the architecture can evolve toward &lt;strong&gt;local inference and self-hosted models&lt;/strong&gt; instead of being permanently tied to one closed AI API.&lt;/p&gt;




&lt;h1&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h1&gt;

&lt;p&gt;Dockerization is one of those problems where there are many different valid solutions depending on the project.&lt;/p&gt;

&lt;p&gt;A Python application, a Node.js application, a Go service, and a project with native system dependencies all require different decisions.&lt;/p&gt;

&lt;p&gt;Open AI models make it possible to build tools around that reasoning without treating a single proprietary AI provider as the foundation of the product.&lt;/p&gt;

&lt;p&gt;For DockForge, open innovation matters in a few important ways.&lt;/p&gt;

&lt;h3&gt;
  
  
  Model freedom
&lt;/h3&gt;

&lt;p&gt;The application isn't fundamentally tied to one model.&lt;/p&gt;

&lt;p&gt;As better open-weight coding models become available, DockForge can adopt them without rebuilding the entire product around a proprietary API.&lt;/p&gt;

&lt;h3&gt;
  
  
  Local and self-hosted possibilities
&lt;/h3&gt;

&lt;p&gt;The same architecture can evolve toward local inference and self-hosted models.&lt;/p&gt;

&lt;p&gt;That opens the door to using DockForge in environments where source code cannot or should not be sent to an external AI provider.&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>weekendchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>Build an AI email summarizer with event planning</title>
      <dc:creator>Gavindu achintha</dc:creator>
      <pubDate>Sat, 13 Sep 2025 15:56:16 +0000</pubDate>
      <link>https://dev.to/gavinduachintha/build-an-ai-email-summarizer-with-event-planning-4e1e</link>
      <guid>https://dev.to/gavinduachintha/build-an-ai-email-summarizer-with-event-planning-4e1e</guid>
      <description>&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fmgrl0ow86hckdzd7wvg3.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fmgrl0ow86hckdzd7wvg3.png" alt=" " width="799" height="382"&gt;&lt;/a&gt;&lt;br&gt;
Built a full-stack email dashboard with Gmail API integration and Google Calendar event planning using React, Node.js,OpenRouter and Kiro AI to accelerate development for the Code with Kiro hackathon.&lt;/p&gt;

</description>
      <category>kiro</category>
      <category>webdev</category>
      <category>javascript</category>
      <category>ai</category>
    </item>
  </channel>
</rss>
