DEV Community: Nishant Gaurav

GitGhoul: The Haunted Codebase & Zombie Code Exorcist with Google Gemma 2

Nishant Gaurav — Wed, 28 Oct 2026 01:42:49 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge: Week 4

🕷️ What I Built: GitGhoul

Every long-lived codebase harbors restless spirits:

🧟 Zombie Code: Dead functions, obsolete endpoints, and unreachable returns lingering after rushed merges.
⚰️ Comment-Cloth Corpses: Old implementations commented out instead of deleted because developers fear losing history.
🕯️ Spectral TODOs: Ancient comments from 2021 like // TODO: Temporary patch until next sprint that haunt PRs for years.
🌀 Cursed Complexity: Deep callback abysses and spaghetti logic where execution paths lose their way.

Standard static analysis tools output sterile, robotic warning tables that developers ignore.

GitGhoul turns technical debt into an engaging Halloween ritual:

Interactive Graveyard UI: Visualizes your codebase as an atmospheric cemetery where files are plotted as weathered tombstones and possession severity scores.
Heuristic Ghoul Detection: Scans code for dead branches, post-return statements, embalmed commented functions, and ancient TODO markers.
Local AI Exorcisms powered by Google Gemma 2: Conducts an on-device exorcism using Google's open-weight Gemma 2 (via Ollama). Gemma purifies the code, writes a witty, spooky Repose Blessing in the spirit of Día de los Muertos, and produces a clean, production-ready unified git diff ready for pull requests.

🔒 Why Open Innovation Matters

The Hacktoberfest prompt asks: Why does open innovation matter for what you built?

For source code analysis and automated refactoring, open-weight, on-device AI is essential:

Total Code Sovereignty & Privacy: Source code contains company secrets, proprietary business logic, unreleased algorithms, and API patterns. Uploading company codebases to third-party commercial cloud AI APIs creates unacceptable IP and compliance risks. With Gemma 2 running 100% locally via Ollama, no proprietary code ever leaves your machine.
Runs in the Dark (Completely Offline): GitGhoul operates with the lights (and the Wi-Fi) off. Whether working on an air-gapped internal server or hacking on a train, the exorcist never drops connection.
Zero Token Taxes on Large Refactors: Running static analysis and refactoring across thousands of lines of code would rack up costly token bills on closed cloud APIs. Open-weight models cost $0 in ongoing API fees.

📸 Screenshots & Live UI

GitGhoul visualizing haunted files as tombstones, displaying spectral laments, and banishing code spirits with Google Gemma 2.

🛠️ How I Built It

Frontend & App Architecture: Built on Next.js 16 (App Router + Turbopack) with TypeScript and Tailwind CSS.
Open-Source AI Engine: Google Gemma 2 (gemma2:9.2B) served locally through Ollama (http://localhost:11434), constrained with structured JSON outputs.
Scanner Pipeline: Custom TypeScript AST and heuristic regex analyzer identifying dead branches, orphaned comments, and ancient TODOs.
Gothic Retro Theme: Pixel-accented dark palette, atmospheric grave markers, and celebratory confetti upon completing exorcisms.

💻 Code Repository

GitHub Repository: https://github.com/Imbiber/git-ghoul

🏆 Prize Categories

Ghost in the Machine (Hacktoberfest Week 4 Overall Theme): Directly embodies the theme by turning developer technical debt and dead code into an interactive graveyard ritual and digital Día de los Muertos remembrance.
Best Use of Gemma ($200): Employs Google's open-weight Gemma 2 locally via Ollama for zero-cost, privacy-first code refactoring and diff generation.

SlothSort: The Zero-Effort Clutter Tidier for International Sloth Day

Nishant Gaurav — Thu, 22 Oct 2026 05:20:39 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge: Week 3

🦥 What I Built: SlothSort

Tomorrow is International Sloth Day, and sloths have life figured out: move at your own pace, conserve energy, and let automation do the tedious heavy lifting so you don't have to.

Every developer and knowledge worker knows the dread of opening their ~/Downloads or Desktop folder: a terrifying graveyard of chaotic filenames:

Screen Shot 2026-10-21 at 11.45.12 PM.png
final_pitch_deck_v3_revised_FINAL2.pptx
Receipt_Uber_Eats_Oct14.pdf
IMG_20261019_142311.jpg
contract_nda_mutual_signable_oct.pdf

Cleaning this manually takes endless clicking, dragging, and cognitive fatigue.

SlothSort is a zero-effort desktop and download tidier that channels your inner sloth:

Drop or Paste Clutter: Toss in messy filenames, downloads, or unorganized project files.
Local Gemma 2 Semantic Intelligence: Google's open-weight Gemma 2 (running 100% locally via Ollama) inspects context, categorizes files, and renames them to clean, consistent, ISO/semantic standards (e.g., 2026-10_Invoice_AcmeCorp.pdf).
Lazy Voice Debrief with ElevenLabs: Listen to a soothing, witty sloth audio debrief celebrating how much effort you saved today.
1-Click Execution Script: Provides an auto-generated bash script (organize.sh) so you can execute the sorting on your machine in one terminal command.

🎥 Video Demo & Walkthrough

▶️ Watch the full walkthrough on YouTube: https://youtu.be/TANmvRYIfnw

📸 Screenshots

SlothSort running 100% on-device with Gemma 2 sorting queue, ElevenLabs voice debrief player, and 1-click terminal execution script.

🔒 Why Open Innovation Matters

The Hacktoberfest prompt asks: Why does open innovation matter for what you built?

For personal file management, open-weight, on-device AI is not just a nice-to-have; it's a non-negotiable requirement:

Absolute Privacy on Sensitive Files: Desktop clutter is full of private data — bank statements, unreleased source code, medical invoices, passwords, and client NDAs. Sending this metadata to commercial cloud LLMs creates catastrophic privacy risks. With Gemma 2 running locally via Ollama, zero bytes ever leave your laptop.
$0 API Costs Forever: Commercial cloud LLM tokens add up quickly when organizing hundreds of files. With open-weight models, you set it up once and clean your folders an unlimited number of times for free.
Offline Resilience: Whether you're on a long flight without Wi-Fi or working in a remote cabin, your sloth organizer works completely offline.

🛠️ How I Built It

Framework: Next.js 16 (Turbopack, App Router) with Tailwind CSS and TypeScript.
Open-Source AI Core: Google Gemma 2 (gemma2:9.2B) served locally through Ollama (http://localhost:11434), producing structured JSON categorization plans.
Voice AI: ElevenLabs Multilingual v2 generating soothing, sloth-themed audio debriefs.
Automation Pipeline: Client-side interactive dashboard + auto-generated Bash script builder.

💻 Code Repository

GitHub Repository: https://github.com/Imbiber/sloth-sort

🤖 Development Session

Show your work: I built this project with AI agent pairing. You can view the agent session transcript here:

Building SlothSort for HF26 Week 3 Gemma 2 (Ollama)

You

User initiated Hacktoberfest Open-Source AI Challenge: Week 3 'Slow and Steady' build. We architected SlothSort: a zero-effort local file organizer using Google Gemma 2 (Ollama) and ElevenLabs voice AI.

Agent

Built complete Next.js app with full local Gemma 2 classification, automatic semantic renaming, lazy time calculation, 1-click bash automation, and ElevenLabs voice debriefing.

🏆 Prize Categories

Best Use of Gemma ($200): Local inference using Google's open-weight Gemma 2 model via Ollama to intelligently classify, semantically rename, and compute lazy metrics for messy desktop clutter without external cloud APIs.
Best Use of ElevenLabs ($100): Integrated ElevenLabs Multilingual v2 to generate soothing, witty voice debriefs celebrating sloth-level productivity.

From Raw Notes to First Sale: The Complete Guide to Building a Privacy-First E-Commerce AI Copywriter with Google Gemma 2 & ElevenLabs

Nishant Gaurav — Thu, 15 Oct 2026 22:06:37 +0000

From Raw Notes to First Sale: The Complete Guide to Building a Privacy-First E-Commerce AI Copywriter with Google Gemma 2 & ElevenLabs

Submitted for the **Hacktoberfest Open-Source AI Challenge: Week 2 — Side Hustle.

💡 The Problem: The Copywriting Chokehold on Solo Side Hustlers

If you run an e-commerce side hustle—selling vintage decor, handcrafted leather goods, custom mechanical keyboards, or 3D-printed workshop tools—you know the feeling: creating the item is the fun part. Writing listings for multiple marketplaces is painful.

Each marketplace speaks its own dialect:

🌿 Etsy expects sensory, artisanal storytelling, tactile material descriptions, and long-tail comma-separated search tags.
📦 Amazon demands punchy, keyword-dense bullet points with zero filler.
🏷️ eBay demands condition grading clarity, straight-to-the-point specs, and reassuring return/shipping guarantees.

┌─────────────────┐       ┌────────────────────────────────────────────────────────┐
│  Raw Item Notes │  ──►  │ How do I write 3 different formats for Etsy, Amazon,   │
│  "Leather card  │       │ & eBay without paying $30/mo or giving up my catalog?  │
│   holder wallet"│       └────────────────────────────────────────────────────────┘
└─────────────────┘

Why Existing Cloud Solutions Fail Side-Hustlers

The Subscription Tax: $20–$50/month per-seat fees or unpredictable OpenAI API token bills eat away at thin profit margins on $15–$30 items.
Catalog Privacy & Leakage: Side-hustlers often draft listings weeks before a collection drop. Unreleased designs, wholesale margins, and proprietary supplier details get fed into commercial cloud LLMs.
The "Silent Copy" Trap: When writing copy silently in your head, awkward phrasing slips through. You only discover your listing sounds stiff and robotic after a buyer ignores it.

ListingGenius was built to solve all three issues:

Runs 100% locally using Google's Gemma 2 via Ollama: $0.00 in API costs, total privacy, and full offline capability.
Features instant voice auditions via ElevenLabs Multilingual v2: Hear how your listing sounds to a real shopper before publishing, or export the MP3 audio directly for TikTok Shop, Reels, and Amazon Video listings.

🎥 Video Demo & Walkthrough

Watch ListingGenius in action generating listings with Google Gemma 2 and auditioning voice with ElevenLabs:

📸 The Application: Live Walkthrough

Here is the live generator in action showing the new marketplace card selector and generated listing output:

Source Code: https://github.com/Imbiber/listing-genius
Interactive Agent Session: Building ListingGenius on DEV

🛠️ Step-by-Step Walkthrough: How It Works

Step 1: Platform Selection & Raw Bullet Input

Instead of struggling over sentence structure, sellers select their target marketplace (Etsy, eBay, or Amazon) and paste raw bullet points or photos:

Product: Minimalist front pocket cardholder wallet
Material: Full grain Italian vegetable-tanned leather
Capacity: Holds 8 cards + cash slot
Features: Hand-stitched with waxed linen thread, RFID blocking layer
Who it's for: Everyday carry enthusiasts, minimalist gift idea

Step 2: Local Gemma 2 Inference ($0 Cost & Instant JSON)

When the user clicks "✨ Generate Listing", the request hits the local Ollama instance running Google's gemma2:latest.

Gemma 2 evaluates the marketplace prompt template and returns structured JSON containing:

Title: Search-ranked and formatted to character limits.
Description: Polished body copy tailored for the selected audience.
Tags: High-volume marketplace search tags.
Call to Action: Direct closing pitch to drive immediate checkouts.

Step 3: Audio Auditioning with ElevenLabs

Once generated, the seller clicks "Read Aloud".

ListingGenius calls ElevenLabs' eleven_multilingual_v2 model with natural speech synthesis:

Why This Changes the Game:

Catches awkward cadence: When spoken aloud, run-on sentences and robotic marketing buzzwords immediately stand out.
Multichannel Video Commerce: Sellers can export the audio file with one click and overlay it on product video shorts for TikTok Shop, Pinterest Idea Pins, or Instagram Reels—turning text into a voiceover video ad in 30 seconds.
Accessibility: Provides immediate audio descriptions for visually impaired buyers.

🏛️ Local Architecture: How We Built It

┌──────────────────────────────────────────────────────────────────┐
│                   Next.js 16 (App Router & Turbopack)            │
│   • Marketplace Switcher (Etsy / Amazon / eBay)                  │
│   • Form Input & One-Click Copy System                           │
│   • Audio Waveform & MP3 Streamer                                │
└─────────────────────────────────┬────────────────────────────────┘
                                  │
                  ┌───────────────┴───────────────┐
                  ▼                               ▼
  ┌──────────────────────────────┐┌──────────────────────────────┐
  │      Google Gemma 2          ││    ElevenLabs TTS Engine     │
  │  (Local Ollama Inference)    ││   (eleven_multilingual_v2)   │
  │                              ││                              │
  │ • Runs on localhost:11434    ││ • Lifelike voice synthesis   │
  │ • Zero API Token Bills       ││ • Audio auditioning          │
  │ • 100% Private & Offline     ││ • TikTok/Reels exportable    │
  │ • Structured JSON Schema     ││ • Natural phrasing validator │
  └──────────────────────────────┘└──────────────────────────────┘

The Prompt Architecture

To ensure Gemma 2 produces predictable, production-ready listings, we constructed platform-tailored system instructions that constrain output to verified JSON schemas:

// lib/prompts.ts
export function getPlatformPrompt(platform: Platform, notes: string) {
  const constraints = {
    etsy: "Craft an emotional, artisanal narrative. Emphasize craftsmanship and materials. Provide 13 search tags.",
    amazon: "Create 5 structured bullet points highlighting benefits, dimensions, and materials. Include high-intent SEO keywords.",
    ebay: "Prioritize clear condition grading, precise technical specs, and reassuring buyer policies (shipping/returns)."
  };

  return {
    system: `You are an elite e-commerce copywriter. Output MUST be valid JSON with keys: title, description, tags, callToAction.`,
    user: `Target Platform: ${platform}. Specifications: ${constraints[platform]}. Product Notes: ${notes}`
  };
}

⚡ Quickstart: Run It Locally in 3 Steps

Want to run ListingGenius on your laptop for your own products?

1. Install Ollama and Pull Gemma 2

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull Google's open-weight model (5.4 GB)
ollama pull gemma2

2. Clone the Repository & Install Dependencies

git clone https://github.com/Imbiber/listing-genius.git
cd listing-genius
npm install

3. Add Keys & Start

Create .env.local:

ELEVENLABS_API_KEY=your_key_here
OLLAMA_BASE_URL=http://localhost:11434

Start development server:

npm run dev

Open http://localhost:3000 and start generating!

🏆 Prize Categories

Theme: Side Hustle (Hacktoberfest Week 2): Solves the copywriting bottleneck for micro-sellers with local AI.
Best Use of Gemma ($200): Runs Google Gemma 2 locally for zero-token-cost, private inference.
Best Use of ElevenLabs ($100): Employs eleven_multilingual_v2 for voice auditioning and video listing voiceovers.

🤖 The Agent Session: Show Your Work

ListingGenius was planned, built, and debugged collaboratively using an AI coding agent. We configured local Ollama fallback pipelines, fixed deprecated ElevenLabs model endpoints, and styled the UI.

You can inspect the entire prompt history, tool calls, and code diffs in the interactive transcript on DEV:

Post not found or has been removed.

💬 Community Feedback

Are you a maker, vintage curator, or digital seller? What marketplace formats would you like added next (Shopify, Depop, Poshmark)? Let me know in the comments below!