<?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: Abdul Khadar</title>
    <description>The latest articles on DEV Community by Abdul Khadar (@abdulkhadhar).</description>
    <link>https://dev.to/abdulkhadhar</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%2F497893%2F0dbea095-65b0-404b-b553-924f2af4ec03.jpg</url>
      <title>DEV Community: Abdul Khadar</title>
      <link>https://dev.to/abdulkhadhar</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/abdulkhadhar"/>
    <language>en</language>
    <item>
      <title>Memory Séance Lite: A Private, Local AI Ofrenda for the Stories We Almost Forget</title>
      <dc:creator>Abdul Khadar</dc:creator>
      <pubDate>Thu, 29 Oct 2026 15:07:07 +0000</pubDate>
      <link>https://dev.to/abdulkhadhar/memory-seance-lite-a-private-local-ai-ofrenda-for-the-stories-we-almost-forget-58p</link>
      <guid>https://dev.to/abdulkhadhar/memory-seance-lite-a-private-local-ai-ofrenda-for-the-stories-we-almost-forget-58p</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;Every family has a ghost like this: a grandmother who named every tomato plant in her garden, an uncle's terrible jokes, a side project that stopped at version 0.3 and has been rotting in a private repo ever since. They are not dramatic stories. They are small, specific, and they disappear the moment the last person who remembers them stops talking.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Memory Séance Lite&lt;/strong&gt; is a small app for keeping those stories. You write short memories about a person, a pet, a place, or an abandoned software project. They are saved in a SQLite file on your own computer. Then a small open-weight language model, running locally through Ollama, does two things:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Writes an ofrenda.&lt;/strong&gt; Pick who or what to remember, and the model writes a short remembrance paragraph, three "things remembered", and a closing line inspired by Día de los Muertos. Only facts from your saved memories are allowed in.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Runs a memory séance.&lt;/strong&gt; Ask a question and the app retrieves the most relevant memories, has the model answer from them alone, and shows you exactly which memories it used. If nothing you saved is relevant, it says it can't answer.&lt;/li&gt;
&lt;/ol&gt;

&lt;h3&gt;
  
  
  How I read the theme
&lt;/h3&gt;

&lt;p&gt;"Ghost in the Machine" can mean something spooky, or something that honors the past. I went for the second, and I tried to let both holidays meet in the middle. Halloween gives the season, the dark background, the candlelight. Día de los Muertos gives the &lt;em&gt;spirit&lt;/em&gt;: remembrance as something warm, communal and colorful, where the dead are welcomed rather than feared. The interface is deliberately dark but gentle, with marigold and candle tones and a strip of papel-picado-style color at the top.&lt;/p&gt;

&lt;p&gt;The "ghost" is not a spirit I pretend to summon. It is the &lt;strong&gt;trace a person or a project leaves behind&lt;/strong&gt;, and the "machine" is a small model that holds that trace for you without making anything up.&lt;/p&gt;

&lt;h3&gt;
  
  
  The line I would not cross
&lt;/h3&gt;

&lt;p&gt;A "séance" app could easily slide into pretending to speak &lt;em&gt;as&lt;/em&gt; someone who has died. I didn't want to build that. It is unkind to people who are grieving, and small models are terrible at it. So:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The narrator always speaks in the &lt;strong&gt;third person&lt;/strong&gt;. It tells you what your memories say; it never says "I" as the person.&lt;/li&gt;
&lt;li&gt;A disclaimer is on screen at all times: &lt;em&gt;the séance is a creative memorial conversation, not communication with a deceased person.&lt;/em&gt;
&lt;/li&gt;
&lt;li&gt;Every answer lists the memories that produced it. If something isn't in your memories, the app says so instead of inventing it.&lt;/li&gt;
&lt;/ul&gt;

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

&lt;p&gt;  &lt;iframe src="https://www.youtube.com/embed/PM4JgmZCKf8" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;p&gt;&lt;em&gt;A 60-second walkthrough: viewing memories, lighting the candle for an ofrenda with Gemma 3 1B, asking the séance a grounded question with citations, and triggering an honest refusal when memories cannot answer.&lt;/em&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%2Fjv7tf5qbhfc9911py00l.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%2Fjv7tf5qbhfc9911py00l.png" alt="Memories View" width="800" height="396"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;To run it yourself (Python 3.10+ and &lt;a href="https://ollama.com/download" rel="noopener noreferrer"&gt;Ollama&lt;/a&gt; required):&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;git clone https://github.com/AbdulKhadhar/memory-seance-lite &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="nb"&gt;cd &lt;/span&gt;memory-seance-lite
./scripts/setup.sh            &lt;span class="c"&gt;# Windows: .\scripts\setup.ps1   (add --low / -Low for qwen3:0.6b)&lt;/span&gt;
python &lt;span class="nt"&gt;-m&lt;/span&gt; app.main            &lt;span class="c"&gt;# or .venv/bin/python -m app.main&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;


&lt;p&gt;Open &lt;code&gt;http://127.0.0.1:8000&lt;/code&gt;. After the one-time download of Ollama and the model, &lt;strong&gt;the app works with no internet connection at all.&lt;/strong&gt; There is also &lt;code&gt;python scripts/load_sample.py&lt;/code&gt;, which loads fictional sample memories so you can try it without writing your own.&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/AbdulKhadhar" rel="noopener noreferrer"&gt;
        AbdulKhadhar
      &lt;/a&gt; / &lt;a href="https://github.com/AbdulKhadhar/memory-seance-lite" rel="noopener noreferrer"&gt;
        memory-seance-lite
      &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;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;Memory Séance Lite&lt;/h1&gt;
&lt;/div&gt;

&lt;p&gt;A private, local-first memorial app: save short memories of a person, pet, place, or abandoned project, and let a small open-weight model on your own computer write an ofrenda and answer questions, using only what you wrote.&lt;/p&gt;

&lt;p&gt;&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/2572e00a3e77288a0c00a8e4ce2f012d92788a36e861449f58fd95bc65bce498/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4d49542d663461353262"&gt;&lt;img src="https://camo.githubusercontent.com/2572e00a3e77288a0c00a8e4ce2f012d92788a36e861449f58fd95bc65bce498/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4d49542d663461353262" alt="License: MIT"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/3f9ca96212106ae38115a4c9008785a3b562655e8acf5be01ebfabf43bc7a317/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e31302532422d333737366162"&gt;&lt;img src="https://camo.githubusercontent.com/3f9ca96212106ae38115a4c9008785a3b562655e8acf5be01ebfabf43bc7a317/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e31302532422d333737366162" alt="Python 3.10+"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/d4873f832b8cfcedce0fbb9dc320b3f6079a08cd5e6ad33bd1a07e15b1cc6c35/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f72756e74696d652d4f6c6c616d612d626c61636b"&gt;&lt;img src="https://camo.githubusercontent.com/d4873f832b8cfcedce0fbb9dc320b3f6079a08cd5e6ad33bd1a07e15b1cc6c35/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f72756e74696d652d4f6c6c616d612d626c61636b" alt="Ollama"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/9dd7933ea1c9d23ee00c338dc1801b111187ea90f73a8cf0891a4107d4c7226b/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6f70656e253230736f757263652d7965732d627269676874677265656e"&gt;&lt;img src="https://camo.githubusercontent.com/9dd7933ea1c9d23ee00c338dc1801b111187ea90f73a8cf0891a4107d4c7226b/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6f70656e253230736f757263652d7965732d627269676874677265656e" alt="Open source"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/428782a17cf093c68f8c4174eb84a2c2b173ecbfa4d3d37f53b90e38c115164e/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f636c6f75642d6e6f6e652d6c6967687467726579"&gt;&lt;img src="https://camo.githubusercontent.com/428782a17cf093c68f8c4174eb84a2c2b173ecbfa4d3d37f53b90e38c115164e/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f636c6f75642d6e6f6e652d6c6967687467726579" alt="No cloud"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;The séance is a creative memorial conversation. It is not communication with a deceased person.&lt;/strong&gt; The narrator is a small language model that only retells the memories you wrote.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Why this exists&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Grief, nostalgia and half-forgotten projects are some of the most private things we write down, and they are exactly what you should not upload to a hosted AI service. Memory Séance Lite keeps the whole loop on your computer: SQLite for storage, Ollama for inference, and a model small enough for an old laptop.&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;The "Ghost in the Machine" theme&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Built for the DEV Hacktoberfest Open-Source AI Challenge, Week 4: &lt;em&gt;Ghost in the Machine&lt;/em&gt;. The "ghost" here…&lt;/p&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/AbdulKhadhar/memory-seance-lite" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;A quick tour of the repository:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Path&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;app/routes.py&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;The API, and the refusal logic for the séance&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;app/retrieval.py&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Tokenizing the question and ranking memories&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;app/database.py&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;SQLite schema, FTS5 index, full wipe&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;app/llm.py&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;A thin Ollama client with typed errors, and the ofrenda JSON validator&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;app/prompts.py&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Short, grounded system prompts&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;code&gt;static/&lt;/code&gt;, &lt;code&gt;templates/&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;Vanilla HTML/CSS/JS, no build step, no &lt;code&gt;node_modules&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;tests/&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;27 pytest tests; every LLM call is mocked&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The whole backend has three runtime dependencies: &lt;code&gt;fastapi&lt;/code&gt;, &lt;code&gt;uvicorn&lt;/code&gt; and &lt;code&gt;httpx&lt;/code&gt;.&lt;/p&gt;

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

&lt;h3&gt;
  
  
  The shape of the system
&lt;/h3&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%2F9hrgpzjmmqs18pn1y3o1.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%2F9hrgpzjmmqs18pn1y3o1.png" alt="System Architecture" width="798" height="138"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Open models: Gemma 3 1B, with Qwen3 0.6B as a fallback
&lt;/h3&gt;

&lt;p&gt;The whole point was that a person with an old laptop could run this. &lt;strong&gt;Gemma 3 1B&lt;/strong&gt; is the default: small enough to download and run on modest hardware, and good enough to retell a handful of short memories in plain language. For machines that are really struggling, there is a one-line switch (in the Settings tab, in &lt;code&gt;.env&lt;/code&gt;, or &lt;code&gt;./scripts/setup.sh --low&lt;/code&gt;) to &lt;strong&gt;Qwen3 0.6B&lt;/strong&gt;. The app sends &lt;code&gt;think: false&lt;/code&gt; to Qwen3 so it skips its reasoning trace, which saves time on weak hardware, and it strips any stray &lt;code&gt;&amp;lt;think&amp;gt;&lt;/code&gt; blocks anyway.&lt;/p&gt;

&lt;p&gt;Because the model is "just a name in a setting", you can also point it at any other model you have installed in Ollama. I did not want the project married to one model.&lt;/p&gt;

&lt;h3&gt;
  
  
  Retrieval without embeddings
&lt;/h3&gt;

&lt;p&gt;Most "chat with your notes" tutorials reach for an embedding model and a vector database. That is exactly the weight I was trying to avoid. For a few hundred short personal memories, &lt;strong&gt;SQLite's built-in FTS5 full-text search with BM25 ranking&lt;/strong&gt; is plenty:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Title and person/project matches are weighted more heavily than body text.&lt;/li&gt;
&lt;li&gt;A Porter stemmer means "growing" finds "grew"-adjacent forms like "grow" and "grows" (it is keyword-based, so it won't magically bridge synonyms).&lt;/li&gt;
&lt;li&gt;Common stopwords are dropped, so "What did she do?" doesn't match everything.&lt;/li&gt;
&lt;li&gt;If a Python build ships without FTS5, a tiny pure-Python scorer takes over automatically.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Only &lt;strong&gt;3 to 5 memories&lt;/strong&gt; ever go into a prompt, with &lt;code&gt;num_ctx&lt;/code&gt; set to 2048. That keeps RAM and latency low.&lt;/p&gt;

&lt;h3&gt;
  
  
  Making a 1B model behave: grounding by design
&lt;/h3&gt;

&lt;p&gt;A small model will happily hallucinate a heartwarming detail that never happened. In a memorial app, that is not a quirk; it is a harm. So I did not rely on the prompt alone. The séance has three separate gates:&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%2F4jal87ky66zrf8urq1vk.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%2F4jal87ky66zrf8urq1vk.png" alt="Séance Gates" width="599" height="1338"&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%2Fwdcrk8pgit5fzu2y6yuf.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%2Fwdcrk8pgit5fzu2y6yuf.png" alt="Grounded Séance Answer" width="800" height="396"&gt;&lt;/a&gt;&lt;br&gt;
&lt;em&gt;Séance answer with retrieved memories expanded and cited.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;The first two gates are plain code, not prompting, so the model never gets the chance to improvise when there is nothing to ground it. The third gate handles the sneaky case: a memory &lt;em&gt;matched&lt;/em&gt; a keyword but doesn't actually answer the question, and the model is told to reply with a sentinel instead of guessing.&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%2Flpgzs8yeu72gc2525a7i.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%2Flpgzs8yeu72gc2525a7i.png" alt="Honest Refusal" width="800" height="396"&gt;&lt;/a&gt;&lt;br&gt;
&lt;em&gt;Honest refusal: When asked a question outside the saved memories, the app refuses to hallucinate.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Here is the full séance request as a sequence:&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%2F50db9u0iqb1nlvbe7rzi.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%2F50db9u0iqb1nlvbe7rzi.png" alt="Séance Sequence" width="800" height="416"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  The ofrenda, and what happens when the model fumbles
&lt;/h3&gt;

&lt;p&gt;Small models are unreliable at structured output, so the ofrenda has its own safety net:&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%2F5h8da9kfj2sdcv38jqi7.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%2F5h8da9kfj2sdcv38jqi7.png" alt="Ofrenda Generation Flow" width="798" height="144"&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%2F0n0t8abm28vcxpl97ary.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%2F0n0t8abm28vcxpl97ary.png" alt="Generated Ofrenda" width="799" height="446"&gt;&lt;/a&gt;&lt;br&gt;
&lt;em&gt;Generated candle-lit ofrenda card with remembered items and closing line.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;The validator tolerates the usual small-model habits (JSON wrapped in prose or code fences, a string where a list should be) and trims to exactly three remembered items. If it still fails after one retry, the user gets a plain explanation, not a stack trace.&lt;/p&gt;

&lt;h3&gt;
  
  
  Failing gracefully
&lt;/h3&gt;

&lt;p&gt;Each failure has a typed error and a human message:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;What went wrong&lt;/th&gt;
&lt;th&gt;What the user sees&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Ollama isn't running&lt;/td&gt;
&lt;td&gt;"Cannot reach Ollama at ... Start it with &lt;code&gt;ollama serve&lt;/code&gt;" (503)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Model not pulled&lt;/td&gt;
&lt;td&gt;"Model 'gemma3:1b' is not installed. Run: &lt;code&gt;ollama pull gemma3:1b&lt;/code&gt;" (409)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Model too slow&lt;/td&gt;
&lt;td&gt;A timeout message suggesting the smaller Qwen3 model (504)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Unusable output&lt;/td&gt;
&lt;td&gt;"The model did not return a usable ofrenda" (502)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Empty database&lt;/td&gt;
&lt;td&gt;An invitation to add the first memory&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Bad input&lt;/td&gt;
&lt;td&gt;Field-level validation messages (422)&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h3&gt;
  
  
  Frontend, accessibility, and safety
&lt;/h3&gt;

&lt;p&gt;The UI is hand-written HTML, CSS and JavaScript served by the backend: no framework, no build step, no external fonts. It has a skip link, labelled form fields, visible focus rings, live regions for status messages, and respects &lt;code&gt;prefers-reduced-motion&lt;/code&gt;. All user text is inserted with &lt;code&gt;textContent&lt;/code&gt;, and the server sends a strict Content-Security-Policy, so a memory can never inject markup. The server binds to &lt;code&gt;127.0.0.1&lt;/code&gt; so only the local machine can reach it.&lt;/p&gt;

&lt;h3&gt;
  
  
  Tests and CI
&lt;/h3&gt;

&lt;p&gt;There are 27 pytest tests, running in about a second, with a temporary SQLite file and a fake LLM in place of every model call. They cover adding memories, FTS retrieval and its fallback, &lt;strong&gt;refusing a séance question when no relevant memory exists&lt;/strong&gt;, ofrenda parsing and retries, Ollama-down and model-missing errors, settings validation, and &lt;strong&gt;full database deletion&lt;/strong&gt;. GitHub Actions runs lint and tests on Python 3.10 and 3.12.&lt;/p&gt;

&lt;p&gt;I deliberately do &lt;strong&gt;not&lt;/strong&gt; run a real model in CI: hosted runners don't resemble the modest hardware this targets, downloading a model on every run is wasteful, and generated text isn't reproducible. A manual workflow produces a Markdown build report instead, and the real demo is recorded locally.&lt;/p&gt;

&lt;h3&gt;
  
  
  Resource &amp;amp; Disk Optimization Practices and Fallbacks
&lt;/h3&gt;

&lt;p&gt;Because the project is designed for modest personal machines, every layer includes deliberate memory caps, disk protections, and graceful fallbacks:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;SQLite Low-Footprint Tuning&lt;/strong&gt;:

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;PRAGMA journal_mode=WAL&lt;/code&gt; with &lt;code&gt;synchronous=NORMAL&lt;/code&gt;: Concurrent reads with zero lock contention and minimal disk I/O stalls while maintaining atomicity.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;PRAGMA cache_size=-2000&lt;/code&gt;: Hard 2 MB upper bound on page cache to avoid memory leaks during long-running sessions.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;PRAGMA temp_store=MEMORY&lt;/code&gt;: Temp tables and sorters reside purely in RAM to eliminate SSD write wear.&lt;/li&gt;
&lt;li&gt;Immediate &lt;code&gt;VACUUM&lt;/code&gt; compaction on data deletion to guarantee reclaimed disk space.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Model Lifecycle &amp;amp; RAM Management&lt;/strong&gt;:

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;keep_alive="5m"&lt;/code&gt;: Instructs Ollama to cleanly unload weights from RAM/VRAM 5 minutes after the last query, returning resources to the host system.&lt;/li&gt;
&lt;li&gt;Bounded prompt evaluation: &lt;code&gt;num_ctx=2048&lt;/code&gt;, &lt;code&gt;num_predict=400&lt;/code&gt;, and &lt;code&gt;num_batch=128&lt;/code&gt; cap memory consumption during prompt evaluations.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Layered Fallbacks&lt;/strong&gt;:

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;FTS5 to Pure-Python&lt;/strong&gt;: BM25 ranking via SQLite FTS5 when present; if the local SQLite build lacks FTS5, it automatically falls back to an in-memory token-matching and stemming algorithm.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tiny Model Support&lt;/strong&gt;: Seamless compatibility with &lt;code&gt;gemma3:1b&lt;/code&gt; and tiny models like &lt;code&gt;qwen3:0.6b&lt;/code&gt; (with &lt;code&gt;think=False&lt;/code&gt; to skip reasoning traces that bottleneck low-RAM CPUs).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Robust Output Fallback&lt;/strong&gt;: Multi-pass JSON parser that handles markdown backticks, conversational preambles, and retry attempts before surfacing actionable feedback.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Building it: the agent and tooling side
&lt;/h3&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%2Ffrxxx2rmrw5xy7po6zmx.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%2Ffrxxx2rmrw5xy7po6zmx.png" alt="Agent Tooling &amp;amp; CI" width="800" height="162"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;GitHub Actions and Copilot setup:&lt;/strong&gt; CI runs lint and tests on multiple Python versions, a manual workflow publishes a build report, Dependabot keeps dependencies and actions fresh, and &lt;code&gt;.github/copilot-instructions.md&lt;/code&gt; establishes repository-level hard rules (local-first, grounded answers with citations, and mocked LLM tests).&lt;/li&gt;
&lt;/ul&gt;

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

&lt;p&gt;This is the part of the project where "open" is not decoration. Take the open pieces away and the app stops being possible, or stops being &lt;em&gt;right&lt;/em&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%2Fibksu95t0ndmp24i5zw6.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%2Fibksu95t0ndmp24i5zw6.png" alt="Privacy Architecture" width="604" height="601"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Memories remain private and local.&lt;/strong&gt; A memorial is about as personal as text gets: a dead parent's habits, a pet's last winter, a failed project you're a little ashamed of. With a closed API, every one of those sentences would be sent to someone else's server, logged under someone else's retention policy, and governed by someone else's terms. Here there is no server to breach because there is no server. The data is one SQLite file you can open, back up, move or delete.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;No paid API, no keys, no meter.&lt;/strong&gt; There are no API keys, no billing, no rate limits. Nobody has to decide whether a grandmother's tomato story is "worth" a few cents of tokens.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;It works with the lights off.&lt;/strong&gt; After the initial downloads, the app works with no internet. Remembrance shouldn't depend on a Wi-Fi signal.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;You can inspect, modify and swap models.&lt;/strong&gt; The model is a setting. When a better tiny model appears, change one line. If you don't like how the narrator sounds, the prompts are about twenty lines in &lt;code&gt;app/prompts.py&lt;/code&gt;, and you can edit them.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;It isn't locked into a service that might disappear.&lt;/strong&gt; This one matters most to me. A family's ofrenda should outlive any startup. If a hosted service pivots, raises prices, or shuts down, everything stored there goes with it. A SQLite file, a model on disk and MIT-licensed code do not have an end-of-life date that someone else controls.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;It runs on modest hardware.&lt;/strong&gt; The people with the most to remember are not always the people with the best computers. A 1B model on a CPU-only laptop is slower than a hosted giant, but it is &lt;em&gt;theirs&lt;/em&gt;, and that trade is the whole point.&lt;/p&gt;

&lt;p&gt;One honest note on terms: Gemma's weights are open but are released under Google's Gemma Terms of Use rather than an OSI-approved license, while Qwen3 is Apache-2.0. The app itself is MIT, and works with either.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Spirit of It
&lt;/h2&gt;

&lt;p&gt;I'm using a fictional example here, because I wouldn't put anyone's real private memories in a demo. Imagine a user, Lucía, who saves a few small memories about her grandmother, Abuela Marisol:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Tomatoes in coffee cans.&lt;/strong&gt; She grew tomatoes in old coffee cans along the back fence and gave every plant a name. The biggest was called Don Ramón.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Sunday pan dulce.&lt;/strong&gt; Every Sunday she walked to the corner bakery and came back with conchas wrapped in paper. She always saved the pink one for me.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Her radio.&lt;/strong&gt; A yellow radio on the kitchen windowsill, tuned to old boleros, while she shelled beans.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Here is the exact ofrenda generated by Gemma 3 1B on my machine from those three memories:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Remembrance:&lt;/strong&gt; Her radio played familiar boleros, a comforting soundtrack to her quiet life. She always brought me a pink concha, a small token of love and a reminder of her garden. She cherished the tomatoes, nurturing them in those coffee cans, and named the biggest one, Don Ramón.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Things Remembered:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A yellow radio, conchas, tomatoes in coffee cans&lt;/li&gt;
&lt;li&gt;A loving gesture of a pink concha&lt;/li&gt;
&lt;li&gt;A garden of tomatoes named Don Ramón&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;Closing:&lt;/strong&gt; &lt;em&gt;May the spirits of those who loved you find peace.&lt;/em&gt;&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Then Lucía asks, &lt;em&gt;"What did she grow in the garden?"&lt;/em&gt; and the narrator answers from her memories, with the "Memories used" panel open beside it, showing the tomato memory and its text.&lt;/p&gt;

&lt;p&gt;Then she asks something the memories can't answer: &lt;em&gt;"What was her favorite football team?"&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;The app says it can't answer from the memories saved, and suggests adding one.&lt;/p&gt;

&lt;p&gt;That refusal is the moment I care about most. It is not a limitation I'm apologizing for. It's the difference between a tool that helps you hold on to what you actually remember and one that quietly writes fiction about someone you loved. If Lucía later remembers that Abuela cheered for a local team, she can add that memory, and the next answer will have a source.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Your memories stay on your machine.&lt;/em&gt;&lt;/p&gt;

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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Gemma&lt;/strong&gt;: Gemma 3 1B (&lt;code&gt;gemma3:1b&lt;/code&gt;) is the default model, running entirely locally through Ollama.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of GitHub Copilot&lt;/strong&gt;: the project is automated with GitHub Actions (CI matrix, manual report workflow), Dependabot, and repo-level AI assistant instructions.&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>DoseBuddy: A lazy medicine scheduler for Amma (paste once, never think about it again)</title>
      <dc:creator>Abdul Khadar</dc:creator>
      <pubDate>Sun, 25 Oct 2026 18:52:16 +0000</pubDate>
      <link>https://dev.to/abdulkhadhar/dosebuddy-a-lazy-medicine-scheduler-for-amma-paste-once-never-think-about-it-again-59pc</link>
      <guid>https://dev.to/abdulkhadhar/dosebuddy-a-lazy-medicine-scheduler-for-amma-paste-once-never-think-about-it-again-59pc</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week3-2026-10-19"&gt;Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

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

&lt;p&gt;My Amma takes multiple daily prescriptions across different conditions—blood pressure, thyroid, and calcium supplements. Managing distinct pill timings, coordinating meals, and worrying about clashing interactions was a daily cognitive load. &lt;/p&gt;

&lt;p&gt;&lt;strong&gt;DoseBuddy&lt;/strong&gt; is an offline, private medicine companion with one goal: &lt;strong&gt;make the person managing prescriptions do the least possible work.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Instead of manual sorting, you simply paste the doctor's prescription note or instructions once. DoseBuddy handles the rest automatically:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Big-Print Compartment View:&lt;/strong&gt; Organizes medicines cleanly into Morning, Afternoon, and Night compartments designed for easy readability.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deterministic Safety Engine:&lt;/strong&gt; Detects clashes and contraindications (e.g., thyroid tablets clashing with calcium or iron within 4 hours).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Spoken Multilingual Reminders:&lt;/strong&gt; Speaks out reminders in English, Hindi, and Malayalam using the browser's speech synthesis engine.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tap-to-Tick Daily Tracker:&lt;/strong&gt; A straightforward daily checklist showing what was taken and what's next.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Printable Fridge Sheet &amp;amp; WhatsApp Share:&lt;/strong&gt; One click generates an instant printable fridge card or a formatted WhatsApp summary for family.&lt;/li&gt;
&lt;/ul&gt;

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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Watch the walkthrough video (42s):&lt;/strong&gt; &lt;a href="https://github.com/AbdulKhadhar/dosebuddy/raw/main/docs/demo.mp4" rel="noopener noreferrer"&gt;docs/demo.mp4&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&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/AbdulKhadhar" rel="noopener noreferrer"&gt;
        AbdulKhadhar
      &lt;/a&gt; / &lt;a href="https://github.com/AbdulKhadhar/dosebuddy" rel="noopener noreferrer"&gt;
        dosebuddy
      &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;div&gt;
&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;💊 DoseBuddy&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;&lt;strong&gt;Paste what the doctor said. Get a big-print daily medicine card and spoken reminders.&lt;/strong&gt;
Offline. Private. Free.&lt;/p&gt;
&lt;p&gt;&lt;a href="https://github.com/AbdulKhadhar/dosebuddy/actions/workflows/ci.yml" rel="noopener noreferrer"&gt;&lt;img src="https://github.com/AbdulKhadhar/dosebuddy/actions/workflows/ci.yml/badge.svg" alt="CI"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/395b010f1304c70beb9c30ae0596eb2320909db42b42b60c7d070b0ed625169a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e392532422d333737364142"&gt;&lt;img src="https://camo.githubusercontent.com/395b010f1304c70beb9c30ae0596eb2320909db42b42b60c7d070b0ed625169a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e392532422d333737364142" alt="Python"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/8bc4f310738608ce17270bec91e541e67c4d64eadab34975080eead0c6757932/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f646570656e64656e636965732d6e6f6e652d326561343466"&gt;&lt;img src="https://camo.githubusercontent.com/8bc4f310738608ce17270bec91e541e67c4d64eadab34975080eead0c6757932/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f646570656e64656e636965732d6e6f6e652d326561343466" alt="Dependencies"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/b8cadaa967891081f8f165695470689986c028821dd8a040132f6e661795dc0d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4d49542d626c7565"&gt;&lt;img src="https://camo.githubusercontent.com/b8cadaa967891081f8f165695470689986c028821dd8a040132f6e661795dc0d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4d49542d626c7565" alt="License"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;a href="https://github.com/user-attachments/assets/demo.mp4" rel="noopener noreferrer"&gt;https://github.com/user-attachments/assets/demo.mp4&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;a href="https://github.com/AbdulKhadhar/dosebuddy/raw/main/docs/demo.mp4" rel="noopener noreferrer"&gt;https://github.com/AbdulKhadhar/dosebuddy/raw/main/docs/demo.mp4&lt;/a&gt;&lt;/p&gt;



&lt;/div&gt;

&lt;blockquote&gt;
&lt;p&gt;Not medical advice. DoseBuddy does not replace a doctor or pharmacist. Its interaction list is short and hand-written, so always confirm with a pharmacist.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Why&lt;/h2&gt;
&lt;/div&gt;

&lt;p&gt;Prescriptions are often written in English or medical shorthand, and several medicines taken at once are easy to mix up. DoseBuddy turns a typed note into a schedule one person can read at a glance, and keeps their health data on their own computer.&lt;/p&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Features&lt;/h2&gt;

&lt;/div&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Paste once, get the plan.&lt;/strong&gt; Morning, afternoon and night compartments in large print.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Safety checks that are not AI.&lt;/strong&gt; Plain, testable Python flags clashes such as thyroid with calcium or iron, duplicates, and missing doses or times.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Review before trust.&lt;/strong&gt; Every extracted medicine can be edited before the plan is made.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tick off doses.&lt;/strong&gt; Tap a medicine when taken…&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/AbdulKhadhar/dosebuddy" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;&lt;em&gt;(Live local server: &lt;a href="http://127.0.0.1:8000" rel="noopener noreferrer"&gt;http://127.0.0.1:8000&lt;/a&gt; — works offline with zero installation steps).&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;You can run it in seconds:&lt;br&gt;
&lt;/p&gt;
&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;git clone https://github.com/AbdulKhadhar/dosebuddy.git
&lt;span class="nb"&gt;cd &lt;/span&gt;dosebuddy
python &lt;span class="nt"&gt;-m&lt;/span&gt; dosebuddy
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&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/AbdulKhadhar" rel="noopener noreferrer"&gt;
        AbdulKhadhar
      &lt;/a&gt; / &lt;a href="https://github.com/AbdulKhadhar/dosebuddy" rel="noopener noreferrer"&gt;
        dosebuddy
      &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;div&gt;
&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;💊 DoseBuddy&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;&lt;strong&gt;Paste what the doctor said. Get a big-print daily medicine card and spoken reminders.&lt;/strong&gt;
Offline. Private. Free.&lt;/p&gt;
&lt;p&gt;&lt;a href="https://github.com/AbdulKhadhar/dosebuddy/actions/workflows/ci.yml" rel="noopener noreferrer"&gt;&lt;img src="https://github.com/AbdulKhadhar/dosebuddy/actions/workflows/ci.yml/badge.svg" alt="CI"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/395b010f1304c70beb9c30ae0596eb2320909db42b42b60c7d070b0ed625169a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e392532422d333737364142"&gt;&lt;img src="https://camo.githubusercontent.com/395b010f1304c70beb9c30ae0596eb2320909db42b42b60c7d070b0ed625169a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e392532422d333737364142" alt="Python"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/8bc4f310738608ce17270bec91e541e67c4d64eadab34975080eead0c6757932/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f646570656e64656e636965732d6e6f6e652d326561343466"&gt;&lt;img src="https://camo.githubusercontent.com/8bc4f310738608ce17270bec91e541e67c4d64eadab34975080eead0c6757932/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f646570656e64656e636965732d6e6f6e652d326561343466" alt="Dependencies"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/b8cadaa967891081f8f165695470689986c028821dd8a040132f6e661795dc0d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4d49542d626c7565"&gt;&lt;img src="https://camo.githubusercontent.com/b8cadaa967891081f8f165695470689986c028821dd8a040132f6e661795dc0d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4d49542d626c7565" alt="License"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;a href="https://github.com/user-attachments/assets/demo.mp4" rel="noopener noreferrer"&gt;https://github.com/user-attachments/assets/demo.mp4&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;a href="https://github.com/AbdulKhadhar/dosebuddy/raw/main/docs/demo.mp4" rel="noopener noreferrer"&gt;https://github.com/AbdulKhadhar/dosebuddy/raw/main/docs/demo.mp4&lt;/a&gt;&lt;/p&gt;

&lt;/div&gt;
&lt;blockquote&gt;
&lt;p&gt;Not medical advice. DoseBuddy does not replace a doctor or pharmacist. Its interaction list is short and hand-written, so always confirm with a pharmacist.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Why&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Prescriptions are often written in English or medical shorthand, and several medicines taken at once are easy to mix up. DoseBuddy turns a typed note into a schedule one person can read at a glance, and keeps their health data on their own computer.&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Features&lt;/h2&gt;

&lt;/div&gt;
&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Paste once, get the plan.&lt;/strong&gt; Morning, afternoon and night compartments in large print.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Safety checks that are not AI.&lt;/strong&gt; Plain, testable Python flags clashes such as thyroid with calcium or iron, duplicates, and missing doses or times.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Review before trust.&lt;/strong&gt; Every extracted medicine can be edited before the plan is made.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tick off doses.&lt;/strong&gt; Tap a medicine when taken…&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/AbdulKhadhar/dosebuddy" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;GitHub Repository:&lt;/strong&gt; &lt;a href="https://github.com/AbdulKhadhar/dosebuddy" rel="noopener noreferrer"&gt;AbdulKhadhar/dosebuddy&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;License:&lt;/strong&gt; MIT&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Architecture:&lt;/strong&gt; Zero external dependencies (standard library &lt;code&gt;http.server&lt;/code&gt;, &lt;code&gt;sqlite3&lt;/code&gt;, vanilla HTML/CSS/JS).&lt;/li&gt;
&lt;/ul&gt;

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

&lt;p&gt;The core architectural principle of DoseBuddy is: &lt;strong&gt;Open-Source AI parses the notes, but pure deterministic code makes all safety decisions.&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Local Open-Weight Model (Inference via Ollama):&lt;/strong&gt;

&lt;ul&gt;
&lt;li&gt;Uses local SLMs (defaulting to &lt;code&gt;qwen2.5:1.5b-instruct&lt;/code&gt;, or customizable via &lt;code&gt;DOSEBUDDY_MODEL=gemma2:2b&lt;/code&gt;) running via Ollama to translate unstructured, messy prescription notes into structured JSON.&lt;/li&gt;
&lt;li&gt;Built with a fallback rule-based regex parser: if Ollama is not running, the application continues to function seamlessly without breaking.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deterministic Safety Rules (&lt;code&gt;rules.py&lt;/code&gt;):&lt;/strong&gt;

&lt;ul&gt;
&lt;li&gt;Contraindications and schedule clashes are evaluated using pure, test-covered Python functions. &lt;/li&gt;
&lt;li&gt;Medicine groups and interaction pairs live in &lt;code&gt;groups.json&lt;/code&gt;, allowing anyone to audit or add medications without writing code.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Review-First UX:&lt;/strong&gt;

&lt;ul&gt;
&lt;li&gt;Before any schedule is saved, the user reviews and edits the extracted items. Human-in-the-loop ensures complete accuracy.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Zero-Cloud Stack:&lt;/strong&gt;

&lt;ul&gt;
&lt;li&gt;Single-file web interface with glassmorphic cards and high-contrast typography, served locally via standard library Python.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;/ol&gt;

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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Medical Privacy by Default:&lt;/strong&gt; Prescriptions and health data should never be sent to third-party proprietary LLM APIs or cloud databases. DoseBuddy keeps all data 100% on the user's local machine.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Zero Cost &amp;amp; Resilience:&lt;/strong&gt; No API subscription tiers or token limits. It functions reliably during power outages and spotty internet connectivity.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Transparent, Auditable Safety:&lt;/strong&gt; In closed models, reasoning is an unpredictable black box. With open innovation, our rule sets and local models can be inspected, vetted, and improved by doctors, pharmacists, and community developers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Model Freedom:&lt;/strong&gt; Users are not vendor-locked; they can switch between lightweight models like Gemma, Qwen, or Llama with a single environment flag.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The Lazy Payoff
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Before:&lt;/strong&gt; ~15–20 minutes of squinting at blister packs, flipping through handwritten doctor notes, cross-checking morning vs. evening doses, and scribbling on pillbox labels.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Now:&lt;/strong&gt; A 5-second paste. DoseBuddy extracts the medicines, flags known clashes, and renders the day's schedule instantly.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Real-World Impact:&lt;/strong&gt; Amma tested the schedule this week—the clear visual compartments and the Malayalam audio prompt meant she never had to ask "did I take the thyroid tablet already?"&lt;/li&gt;
&lt;/ul&gt;

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

&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Slow and Steady (Local First &amp;amp; Privacy)&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Best Use of Open Source AI&lt;/strong&gt;&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;em&gt;Team: Just me, &lt;a href="https://github.com/AbdulKhadhar" rel="noopener noreferrer"&gt;@AbdulKhadhar&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
      <category>python</category>
      <category>opensource</category>
    </item>
    <item>
      <title>Eddit – A Simple, Open-Source Video Editing Tool 🚀</title>
      <dc:creator>Abdul Khadar</dc:creator>
      <pubDate>Sun, 02 Mar 2025 13:03:43 +0000</pubDate>
      <link>https://dev.to/abdulkhadhar/eddit-a-simple-open-source-video-editing-tool-2g20</link>
      <guid>https://dev.to/abdulkhadhar/eddit-a-simple-open-source-video-editing-tool-2g20</guid>
      <description>&lt;p&gt;Do you have basic video editing tasks like splitting long recordings into shorter segments, compressing files, or adding intros? Many available tools are either too complex, expensive, or bloated with features you don’t need.&lt;/p&gt;

&lt;h3&gt;
  
  
  Why I Built Eddit ? 🎬
&lt;/h3&gt;

&lt;p&gt;I often found myself needing a simple way to cut videos into multiple segments, compress them, and add intros without using heavy or complex editing software. Most available tools were either too complicated, expensive, or overkill for these basic tasks.&lt;/p&gt;

&lt;p&gt;One common scenario I faced was splitting long tutorial or online class recordings into shorter, more manageable parts. Uploading lengthy videos to a tutorial app was inefficient, but I couldn't find a simple way to segment them. Additionally, I needed to add a short intro to each video. Since I already had the intro video, I wanted an easy way to assign this task to the client themselves. However, the lack of a simple tool left me stuck. That was the first reason for the journey to Eddit!&lt;/p&gt;

&lt;h3&gt;
  
  
  What is Eddit? 🤔
&lt;/h3&gt;

&lt;p&gt;Eddit is a lightweight, open-source video editing tool designed for quick and easy video modifications. No unnecessary features - just a simple, efficient way to handle your videos.&lt;/p&gt;

&lt;h3&gt;
  
  
  Features ✨
&lt;/h3&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.amazonaws.com%2Fuploads%2Farticles%2Flcagp1zr27tpry7t9l7x.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%2Flcagp1zr27tpry7t9l7x.png" alt="Image description" width="800" height="472"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Cut &amp;amp; Save Segments &lt;/li&gt;
&lt;li&gt;Easily split a video into multiple parts and give each one a custom name.&lt;/li&gt;
&lt;li&gt;Adjust Quality &amp;amp; Compress – Reduce file size without compromising on clarity.&lt;/li&gt;
&lt;li&gt;Add Custom Intros – Assign different intros to different segments, or skip them altogether.&lt;/li&gt;
&lt;li&gt;Batch Processing – Process all segments in one click and save them instantly.&lt;/li&gt;
&lt;li&gt;Simple &amp;amp; Fast – No complex UI, no learning curve, just install and start editing.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Open-Source &amp;amp; Free! 🆓
&lt;/h3&gt;

&lt;p&gt;Eddit is completely free and open-source! You can download, use, and even contribute to the project. It’s currently available for Windows and can be installed easily using a simple Windows installer.&lt;/p&gt;

&lt;p&gt;🔗 GitHub Repo: &lt;a href="https://github.com/AbdulKhadhar/eddit" rel="noopener noreferrer"&gt;https://github.com/AbdulKhadhar/eddit&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Built with Tauri, React &amp;amp; Tailwind 4 🛠️
&lt;/h3&gt;

&lt;p&gt;Eddit is built using Tauri 2 (Rust), React (TypeScript), and Tailwind CSS 4. Tauri allowed me to keep the app lightweight while still delivering powerful functionality, and Tailwind 4 made styling fast and efficient.&lt;/p&gt;

&lt;p&gt;Eddit is built using Tauri 2 (Rust) and React (TypeScript). Tauri allowed me to keep the app lightweight while still delivering powerful functionality.&lt;/p&gt;

&lt;h3&gt;
  
  
  Want to Contribute? 🤝
&lt;/h3&gt;

&lt;p&gt;Let's Connect! 🔗&lt;/p&gt;

&lt;p&gt;If you'd like to connect, feel free to reach out to me on LinkedIn: &lt;a href="https://www.linkedin.com/in/abdul-khadar/" rel="noopener noreferrer"&gt;Abdul Khadar&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Since Eddit is open-source, contributions are always welcome! If you’d like to improve the tool, add features, or fix bugs, feel free to check out the repo, submit a PR, or open an issue.&lt;/p&gt;

&lt;p&gt;Would love to hear your thoughts and feedback! Try it out and let me know what you think. 🚀&lt;/p&gt;

</description>
      <category>programming</category>
      <category>tauri</category>
      <category>opensource</category>
      <category>react</category>
    </item>
  </channel>
</rss>
