<?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: Anshika Pandey</title>
    <description>The latest articles on DEV Community by Anshika Pandey (@anshika_pandey).</description>
    <link>https://dev.to/anshika_pandey</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%2F2880575%2Fac9d9943-8fef-4eda-9a9e-8f425588c63e.png</url>
      <title>DEV Community: Anshika Pandey</title>
      <link>https://dev.to/anshika_pandey</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/anshika_pandey"/>
    <language>en</language>
    <item>
      <title>Slothbox: Gemma tidies my Downloads folder while I sleep, at exactly the speed of a sloth</title>
      <dc:creator>Anshika Pandey</dc:creator>
      <pubDate>Wed, 21 Oct 2026 14:35:14 +0000</pubDate>
      <link>https://dev.to/anshika_pandey/slothbox-gemma-tidies-my-downloads-folder-while-i-sleep-at-exactly-the-speed-of-a-sloth-heg</link>
      <guid>https://dev.to/anshika_pandey/slothbox-gemma-tidies-my-downloads-folder-while-i-sleep-at-exactly-the-speed-of-a-sloth-heg</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;
  
  
  The folder I've been avoiding
&lt;/h2&gt;

&lt;p&gt;My Downloads folder had about 140 loose files in it when I started this. When I sorted them oldest-first, the first sixty told the whole story: forty of them were installers. Four separate IntelliJ installers. Two PostgreSQLs, two HeidiSQLs, two Postmans, two Chromes, three identical copies of &lt;code&gt;SupportAssistLauncher&lt;/code&gt;. Somewhere between them, the important stuff: my Aadhaar and PAN card scans, a tax certificate, a train booking, my resume, and a PDF called &lt;code&gt;train (1).pdf&lt;/code&gt; that turned out to be that train booking.&lt;/p&gt;

&lt;p&gt;Every few months I open this folder, look at it, and close it again. That's the whole relationship.&lt;/p&gt;

&lt;p&gt;Sloths aren't really lazy. They're specialists in doing one thing at a time and never spending energy they don't need to. That felt like exactly the right temperament for this job. The work is boring, nobody needs it done &lt;em&gt;fast&lt;/em&gt;, and it can happen while I'm asleep. So I built &lt;strong&gt;Slothbox&lt;/strong&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  What it does
&lt;/h2&gt;

&lt;p&gt;Slothbox is a single Python script. It looks at every loose file in Downloads, one at a time. For each one, a local Gemma 3 model reads what's inside (the text, the first pages of a PDF, the words in a Word or PowerPoint file, the file list of a zip, or the pixels of a screenshot) and decides two things: which category it belongs in, and what a sensible human would have named it.&lt;/p&gt;

&lt;p&gt;Then it moves the file. Recent files go to &lt;code&gt;Sorted/&amp;lt;category&amp;gt;/&lt;/code&gt;. Anything older than 90 days goes to &lt;code&gt;Archive/&amp;lt;year&amp;gt;/&amp;lt;category&amp;gt;/&lt;/code&gt;. At the end it writes a short morning note. Here's a slice of my first real run:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;  train (1).pdf
    → Archive\2025\personal-docs\train-booking-slip-2025-05.pdf
      (The file contains a train booking slip with passenger details and booking information.)
  npp.8.7.1.Installer.x64.exe
    → Archive\2024\installers\npp-8.7.1-installer.exe
      (The file is an executable installer for Notepad++.)
  Dropbox Integration.pdf
    → Archive\2025\work-docs\dropbox-integration-guide.pdf
      (This document describes how to integrate with the Dropbox API.)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Why this had to be open, local AI
&lt;/h2&gt;

&lt;p&gt;For this project, open models aren't a nice-to-have. They're the only reason the tool can exist.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;My Downloads folder is the most sensitive folder on my computer.&lt;/strong&gt; That's where ID scans, tax forms, bank statements and contracts land, precisely because I never got around to filing them anywhere. An AI tidier is only useful if it reads the contents; &lt;code&gt;train (1).pdf&lt;/code&gt; tells you nothing until you open it. With a hosted model, "tidy my Downloads" really means "upload my Aadhaar card and my tax certificate to a server I don't control, every night, forever." I wasn't willing to do that.&lt;/p&gt;

&lt;p&gt;With Gemma running through Ollama, none of it leaves the laptop. You can switch off Wi-Fi and it keeps working. That's not a privacy policy I have to trust. It's just where the bytes are.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;It costs nothing to leave running.&lt;/strong&gt; A tool you schedule every night for years is the worst possible fit for per-token pricing. The cost of my run was the electricity my laptop was already using.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;It doesn't need to be fast, so a small model is enough.&lt;/strong&gt; Gemma 3 4B averaged &lt;strong&gt;14.8 seconds per file&lt;/strong&gt; on my laptop (Intel Core Ultra 7 255H, 32 GB RAM): 55 files in about 14 minutes. In a chat window, fifteen seconds per answer would be painful. At 3 a.m. it doesn't matter at all. This is where "slow and steady" stops being a theme and becomes the design: once you stop needing the answer &lt;em&gt;now&lt;/em&gt;, the model that fits on your own hardware is the right model.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;I could reach inside the stack.&lt;/strong&gt; Two open features did real work here. Ollama's structured outputs let me pass a JSON schema with a fixed list of 13 categories, so the decoder can't produce a category that isn't on the list. And because the model is a flag (&lt;code&gt;--model&lt;/code&gt;), swapping Gemma sizes is a one-word change.&lt;/p&gt;

&lt;h2&gt;
  
  
  How it works
&lt;/h2&gt;

&lt;p&gt;The whole thing is one file of standard-library Python, plus &lt;code&gt;pypdf&lt;/code&gt; for reading PDFs.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;for each top-level file, oldest first:
    skip hidden files, system files, half-finished downloads
    fingerprint = sha256(size + first 1 MB)
    if fingerprint is cached: reuse the decision
    else:
        read a snippet, or attach the image
        ask Gemma for {category, new_name, reason}, constrained by a JSON schema
        rest a few seconds
    clean_decision(): the code checks the model's work
    move into Sorted/ or Archive/, never overwriting, and journal the move
write a morning note
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It runs at the lowest CPU priority, so it never competes with anything else I'm doing. It never deletes anything (Archive is just a folder), never overwrites (clashes become &lt;code&gt;name-2.exe&lt;/code&gt;), shows every proposed move with Gemma's reason before touching anything (&lt;code&gt;slothbox plan&lt;/code&gt;), and keeps a journal so &lt;code&gt;slothbox undo&lt;/code&gt; can put the last run back exactly as it was.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I learned: the model suggests, the code decides
&lt;/h2&gt;

&lt;p&gt;The interesting part of this project wasn't the first run. It was the three runs after it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Run 1: prompt only.&lt;/strong&gt; Gemma recognized every installer, which was impressive, but the names were inconsistent. It kept the version number for VS Code and dropped it for Notepad++. It renamed &lt;code&gt;HeidiSQL_12.8.0.6908_Setup.exe&lt;/code&gt; to &lt;code&gt;heidisql-setup-2024-11-26.exe&lt;/code&gt;: the version vanished and a date appeared that wasn't anywhere in the file. It had taken the "last modified" date I'd shown it and decided that was the file's date. It also filed &lt;code&gt;debug.log&lt;/code&gt; under installers.&lt;/p&gt;

&lt;p&gt;I added rules to the prompt: keep version numbers with their dots, only use dates that appear inside the file, logs aren't installers. Then I found that half the version problem was my own code. My filename cleaner turned every dot into a dash, so &lt;code&gt;mysql-9.1.0&lt;/code&gt; could never survive no matter what Gemma said. Fixed that too.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Run 2: prompt rules.&lt;/strong&gt; Versions were much better: &lt;code&gt;npp-8.7.1&lt;/code&gt;, &lt;code&gt;mysql-9.1.0&lt;/code&gt;, &lt;code&gt;node-v22.17.1&lt;/code&gt;. But the second run surfaced the scariest bug of the project. One of my tax documents had my PAN number in its filename, and Gemma &lt;em&gt;changed one letter of it&lt;/em&gt;. Imagine &lt;code&gt;ABCDE1234F&lt;/code&gt; becoming &lt;code&gt;ABCTE1234F&lt;/code&gt;. A 4B model isn't copying characters; it's predicting them, and an ID number looks a lot like noise to a language model.&lt;/p&gt;

&lt;p&gt;It also tried to move &lt;code&gt;desktop.ini&lt;/code&gt;, a Windows system file. It renamed a Zoom log to a 70-character string of random letters. And it still put a made-up date on the Slack installer, prompt rule or not.&lt;/p&gt;

&lt;p&gt;That's when the design clicked for me. &lt;strong&gt;A prompt is a suggestion. Rules that must never break belong in code.&lt;/strong&gt; So I wrote &lt;code&gt;clean_decision()&lt;/code&gt;, which checks every answer before anything moves:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;If an ID-like token from the original name (8+ characters mixing letters and digits) shows up in the new name with a character or two changed, put the original back.&lt;/li&gt;
&lt;li&gt;If a date's year appears nowhere in the filename or the contents, remove it.&lt;/li&gt;
&lt;li&gt;If the new name is over 60 characters or contains a long random-looking token, fall back to the original name, cleaned up.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;.exe&lt;/code&gt; and &lt;code&gt;.msi&lt;/code&gt; are always installers, &lt;code&gt;.log&lt;/code&gt; is always data. The model doesn't get a vote.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;desktop.ini&lt;/code&gt; and friends are never touched.
&lt;strong&gt;Run 3: guardrails, and actually reading PDFs.&lt;/strong&gt; I'd also discovered that every PDF reason said something like "the file name suggests…", because &lt;code&gt;pypdf&lt;/code&gt; wasn't installed and Gemma was guessing from names alone. Once it could read the pages, the difference was night and day:&lt;/li&gt;
&lt;/ul&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Before&lt;/th&gt;
&lt;th&gt;After&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;train-1.pdf&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;&lt;code&gt;train-booking-slip-2025-05.pdf&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;carelink-3.pdf&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;&lt;code&gt;carelink-consent-form.pdf&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;anshika-pandey-pdf.pdf&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;&lt;code&gt;anshika-pandey-resume.pdf&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;My favorite moment: Gemma described the Zoom log as "a Zoom installer executable file." Wrong. But the code filed it under &lt;code&gt;data&lt;/code&gt; anyway. The model can be confidently mistaken and the file still ends up in the right place.&lt;/p&gt;

&lt;p&gt;That run also crashed halfway, because a file disappeared from Downloads while Slothbox was working on it. A tool that's slow on purpose has to expect the world to change underneath it, so a vanished file is now just "skipped." And because every decision is cached the moment it's made, the rerun picked up exactly where it stopped. The 28 files it had already read didn't cost a second.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;It's still not perfect, and I think that's honest to say.&lt;/strong&gt; &lt;code&gt;IBALauncher.exe&lt;/code&gt; became &lt;code&gt;ibalancher.exe&lt;/code&gt;, the same dropped-character problem as the PAN number, just harmless this time. A JRE zip picked up "windows-installer" in its name, which it isn't. And my first morning note cheerfully mentioned "a few receipts lurking in there." There were zero receipts. It also said "overnight" at 7:46 p.m. Even when you hand a model the exact facts, it will happily add a few of its own.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;ollama pull gemma3:4b
pip &lt;span class="nb"&gt;install &lt;/span&gt;pypdf
git clone https://github.com/Anshika-019/slothbox &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="nb"&gt;cd &lt;/span&gt;slothbox
python slothbox.py plan &lt;span class="nt"&gt;--limit&lt;/span&gt; 20
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Read what it proposes, then &lt;code&gt;apply&lt;/code&gt; when you trust it. To run it every night at 3 a.m. on Windows:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight batchfile"&gt;&lt;code&gt;&lt;span class="nb"&gt;schtasks&lt;/span&gt; &lt;span class="na"&gt;/create /sc &lt;/span&gt;&lt;span class="kd"&gt;daily&lt;/span&gt; &lt;span class="na"&gt;/st &lt;/span&gt;&lt;span class="m"&gt;03&lt;/span&gt;:00 &lt;span class="na"&gt;/tn &lt;/span&gt;&lt;span class="kd"&gt;slothbox&lt;/span&gt; &lt;span class="na"&gt;/tr &lt;/span&gt;&lt;span class="s2"&gt;"python C:\path\to\slothbox\slothbox.py apply"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;(On macOS or Linux, the README has a cron line and a launchd file.)&lt;/p&gt;

&lt;p&gt;Repo: &lt;a href="https://github.com/Anshika-019/slothbox" rel="noopener noreferrer"&gt;https://github.com/Anshika-019/slothbox&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Built with Gemma 3 (open-weight) and Ollama (open source). Submitting for &lt;strong&gt;Best Use of Gemma&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Sixty files down, seventy-eight to go. Slothbox will get to them tonight. 🦥&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
      <category>opensource</category>
      <category>gemma</category>
    </item>
    <item>
      <title>Listing Writer: An Open-Source AI Tool That Writes Etsy &amp; eBay Listings</title>
      <dc:creator>Anshika Pandey</dc:creator>
      <pubDate>Thu, 15 Oct 2026 09:09:59 +0000</pubDate>
      <link>https://dev.to/anshika_pandey/listing-writer-an-open-source-ai-tool-that-writes-etsy-ebay-listings-3jm4</link>
      <guid>https://dev.to/anshika_pandey/listing-writer-an-open-source-ai-tool-that-writes-etsy-ebay-listings-3jm4</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;p&gt;Writing a good product listing takes a small seller 15-20 minutes per item, and most of them hate doing it. Listing Writer turns a product name and a few details (plus an optional photo) into a ready-to-paste title, description and 13 search tags for Etsy or eBay. It runs on Gemma, Google's open-weight model, so there's no paid API behind it.&lt;/p&gt;

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

&lt;p&gt;Listing Writer is a small web app for people who sell physical products online: Etsy makers, eBay resellers, and people flipping used tech.&lt;/p&gt;

&lt;p&gt;You fill in four things:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Product name&lt;/strong&gt;, like "Refurbished iPhone 13"&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Details&lt;/strong&gt;, like storage, colour, battery health and condition&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Platform&lt;/strong&gt;: Etsy or eBay&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tone&lt;/strong&gt;: Friendly, Professional or Playful&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;You can also add a photo, since Gemma 3 can read images. The app returns a keyword-first title under 140 characters, a short description with a bullet list of specs, and 13 tags sized for Etsy's tag limit.&lt;/p&gt;

&lt;p&gt;The idea is simple on purpose. A seller listing 30 items a week loses hours to writing. This gives them a solid first draft in seconds that they only need to check.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Who would pay?&lt;/strong&gt; Sellers who list a lot of items, like eBay resellers flipping used phones and Etsy shops with big catalogues. The selling point is time: seconds instead of 15-20 minutes per item. &lt;strong&gt;As a seller myself, I'd pay around ₹99 a month for this, or a one-time fee for a version that runs on my own laptop.&lt;/strong&gt;&lt;/p&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%2Fpcyvqjq8o8h5g52cr1tv.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%2Fpcyvqjq8o8h5g52cr1tv.png" alt="Earbuds listing on Etsy after the prompt fix" width="800" height="357"&gt;&lt;/a&gt;&lt;br&gt;
Earbuds on Etsy with a Friendly tone. The title, description bullets and tags all stick to the five details I typed in.&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%2F035ekjnlxq3tknj0ipnv.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%2F035ekjnlxq3tknj0ipnv.png" alt="Refurbished iPhone 13 listing on eBay with a Professional tone" width="800" height="605"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Refurbished iPhone 13 on eBay, Professional tone.&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%2Fpgl9kosrnabdc5srnutg.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%2Fpgl9kosrnabdc5srnutg.png" alt="20000 mAh power bank listing on Etsy" width="800" height="446"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;20000 mAh power bank on Etsy, Friendly tone.&lt;/em&gt;&lt;/p&gt;

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

&lt;p&gt;&lt;strong&gt;&lt;/strong&gt;&lt;/p&gt;&lt;div class="ltag-github-readme-tag"&gt;&lt;strong&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;a href="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" class="article-body-image-wrapper"&gt;&lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;&lt;/a&gt;
      &lt;a href="https://github.com/Anshika-019" rel="noopener noreferrer"&gt;
        Anshika-019
      &lt;/a&gt; / &lt;a href="https://github.com/Anshika-019/listing-writer" rel="noopener noreferrer"&gt;
        listing-writer
      &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;Listing Writer&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;Turn a product name and a few details (plus an optional photo) into a ready-to-paste Etsy or eBay listing: a title, a description and 13 search tags.&lt;/p&gt;

&lt;p&gt;Built for the Hacktoberfest 2026 Open-Source AI Challenge (Week 2: Side Hustle).&lt;/p&gt;

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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Gemma 3 (4B)&lt;/strong&gt;, Google's open-weight model&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ollama&lt;/strong&gt; to serve the model locally&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gradio&lt;/strong&gt; for the web interface&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;No paid API: every listing is generated by an open model running on your own hardware or a free Colab GPU.&lt;/p&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Run it in Google Colab (no install)&lt;/h2&gt;
&lt;/div&gt;

&lt;ol&gt;
&lt;li&gt;Open &lt;code&gt;listing_writer_colab.ipynb&lt;/code&gt; in &lt;a href="https://colab.research.google.com" rel="nofollow noopener noreferrer"&gt;Google Colab&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;Go to &lt;strong&gt;Runtime &amp;gt; Change runtime type &amp;gt; T4 GPU&lt;/strong&gt; and save.&lt;/li&gt;
&lt;li&gt;Run the three cells in order:
&lt;ul&gt;
&lt;li&gt;Cell 1 installs Ollama and the Python libraries.&lt;/li&gt;
&lt;li&gt;Cell 2 starts Ollama and downloads Gemma.&lt;/li&gt;
&lt;li&gt;Cell 3 starts the app.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;Open the &lt;code&gt;gradio.live&lt;/code&gt; link printed by Cell 3.&lt;/li&gt;
&lt;/ol&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Run it on your laptop&lt;/h2&gt;

&lt;/div&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; 1. Install Ollama&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/Anshika-019/listing-writer" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/strong&gt;&lt;/div&gt;&lt;strong&gt;

&lt;/strong&gt;&lt;p&gt;&lt;/p&gt;

&lt;p&gt;The repo has one Colab notebook with three cells: install Ollama and Gradio, start Ollama and pull Gemma, and run the app. Open it in Google Colab, pick a T4 GPU runtime, run the cells in order, and open the &lt;code&gt;gradio.live&lt;/code&gt; link it prints. The same Python code also runs on a laptop with Ollama installed.&lt;/p&gt;

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

&lt;p&gt;The stack is three open-source pieces:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Gemma 3 (4B)&lt;/strong&gt;, Google's open-weight model. It is small enough to run on a free Colab T4 GPU or a normal laptop, and it understands images.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ollama&lt;/strong&gt; to download and serve the model locally with one command (&lt;code&gt;ollama pull gemma3:4b&lt;/code&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gradio&lt;/strong&gt; for the web interface. Its &lt;code&gt;share=True&lt;/code&gt; option gives a public link, so anyone can try the app from their phone.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The app itself is about 90 lines of Python. A system prompt tells Gemma how to write listings and fixes the output format (TITLE, DESCRIPTION, TAGS). Each request sends the seller's details, chosen platform and tone, plus the photo if there is one.&lt;/p&gt;

&lt;p&gt;I didn't want to install anything on my machine, so I ran it all in Google Colab. My first attempt failed with "Failed to connect to Ollama" because the Ollama server wasn't running when the app started. I split the notebook into three cells in a fixed order and added a check that waits until Ollama replies before going on.&lt;/p&gt;

&lt;h3&gt;
  
  
  Testing: The Model Made Things Up, So I Fixed the Prompt
&lt;/h3&gt;

&lt;p&gt;I didn't have a real seller to test with in time, so I tested it myself as a seller, with four tech products: wireless earbuds, a mechanical keyboard, a 20000 mAh power bank and a refurbished iPhone 13.&lt;/p&gt;

&lt;p&gt;The first earbuds listing looked great at a glance, but it invented things I never gave it:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;What I typed&lt;/th&gt;
&lt;th&gt;What the first version wrote&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Type-C charging&lt;/td&gt;
&lt;td&gt;"Type-C &lt;strong&gt;Charging Cable Included&lt;/strong&gt;"&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Touch controls&lt;/td&gt;
&lt;td&gt;"Touch Controls - &lt;strong&gt;Volume, Play/Pause, Next/Previous Track&lt;/strong&gt;"&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;(nothing)&lt;/td&gt;
&lt;td&gt;"&lt;strong&gt;I've been really impressed&lt;/strong&gt; with how smoothly they connect"&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;(nothing)&lt;/td&gt;
&lt;td&gt;An intro line: "Okay, here's a product listing draft..."&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;For a real seller this is dangerous. If the listing says a cable is included and it isn't, the buyer complains or asks for a refund.&lt;/p&gt;

&lt;p&gt;I rewrote the system prompt with strict rules: only use facts the seller gave or that are visible in the photo, never add accessories or specs, never write as if you used the product, and start directly with "TITLE:". After that, the bullet list contained exactly the five details I typed, and the fake review and intro line were gone.&lt;/p&gt;

&lt;p&gt;It still isn't perfect. One tag said "waterproof", but IPX5 only means splash and sweat resistant. A small model still needs a human to check before publishing, and I'd say that clearly to any seller using it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The iPhone and power bank listings were also clean: the iPhone listing kept "minor scratches" and "89% battery health" clearly in the description, which matters a lot for a refurbished item.&lt;/strong&gt;&lt;/p&gt;

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

&lt;p&gt;For a side hustle, the cost per listing decides whether the business works at all.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;No per-request bill.&lt;/strong&gt; With a paid API, every listing costs money, and a seller writing hundreds of drafts pays every time. Gemma runs on hardware you already have, so a draft costs nothing extra.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Seller data can stay private.&lt;/strong&gt; On a laptop with Ollama, a seller's product details, prices and photos never leave their machine. I used Colab for this demo, but the same code runs fully offline.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;No lock-in.&lt;/strong&gt; If a better open model comes out, I change one line (&lt;code&gt;MODEL = ...&lt;/code&gt;). If a provider raises prices or shuts down an API, nothing breaks.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Room to specialise.&lt;/strong&gt; A general API writes generic listings. With open weights, I could fine-tune Gemma on real high-performing listings for a niche like refurbished phones, which a closed API won't let me do.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Where closed models were better: a big hosted model probably wouldn't have invented the cable and controls in the first place. With a 4B model, I had to be much stricter in the prompt. For this use case, that trade was worth it.&lt;/p&gt;

&lt;h2&gt;
  
  
  What's Next
&lt;/h2&gt;

&lt;p&gt;Next I want to:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Put it in front of real sellers and see which listings they actually keep.&lt;/li&gt;
&lt;li&gt;Add a bulk mode that reads a CSV of products and writes all the listings at once.&lt;/li&gt;
&lt;li&gt;Add a check that flags risky words like "waterproof" or "brand new" when the details don't support them.&lt;/li&gt;
&lt;li&gt;Fine-tune Gemma on good listings for one niche, like refurbished phones.&lt;/li&gt;
&lt;/ol&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;: the whole app runs on Gemma 3 (4B), served with Ollama, including reading product photos.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Thanks for reading! Feedback and ideas are very welcome in the comments.&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
      <category>ai</category>
      <category>opensource</category>
    </item>
  </channel>
</rss>
