This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
What I Built
Solo sellers, makers, and side-hustlers — selling on Etsy, eBay, Shopify, or local markets — spend countless tedious hours writing product titles, descriptions, keyword tags, and SEO metadata.
For someone testing a side hustle or making an extra $200–$500 a month, paying a $20–$40/month subscription for a proprietary AI listing SaaS is a massive friction point that directly eats their margins.
Seller AI is an open-source, local-first AI workspace built specifically for small online sellers. It turns raw, bullet-point product notes and photos into marketplace-ready, high-converting product listings in seconds:
- 4-in-1 Marketplace Formats: Generates listings tailored to the strict rules and character limits of Generic e-commerce, Shopify, Etsy, and eBay.
- Alternative Titles & SEO Metadata: Generates multiple catchy title hooks, meta descriptions, and algorithmic search tags.
- Multilingual & Regional Commerce: Supports English, pure Hindi (Devanagari script), and bilingual listings for sellers targeting regional and global buyers.
-
Local Vision Analysis: Capability-aware image analysis for product photos running on open-weight vision models (like
qwen2.5vl:3b) without sending unreleased product photos to third-party clouds. -
Zero Lock-In & Complete Privacy: Client data and product ideas never leave the machine. Listings are stored locally in the browser (
localStorage), with 1-click copy for individual fields, copy-all, and export to TXT, CSV, and JSON.
Demo
deplyed link: https://seller-ai-ldyl.onrender.com
Code
GitHub Repository: https://github.com/Nareshkumawat-star/seller-ai-
How I Built It
- Open-Source AI Stack Seller AI is powered entirely by open-weight models running on local inference through Ollama:
Text & Copywriting: qwen2.5:3b (or llama3.2:3b / mistral:7b) for lightweight, high-efficiency generation.
Vision Extraction: qwen2.5vl:3b for extracting physical attributes, materials, and colors from product photos.
- Architecture & Reliability Patterns Prompt Isolation & Security: User input is strictly isolated inside blocks with system-level instruction guarding against prompt injection. Structured JSON Mode: The model is prompted with format: "json". Responses pass through a robust extraction pipeline that handles markdown code fences, comments, and chatter, followed by Zod runtime schema validation. Language Verification Safeguard: Early tests revealed that smaller models sometimes claim language: "hi" while still returning English body text. We implemented a Unicode script analyzer (lib/ai/language-check.ts) that verifies Devanagari script density, automatically retries if English is detected, and flags unverified outputs. Hardware-Honest Health Check: /api/health queries Ollama's /api/tags and /api/show to verify if the configured models are actually pulled and vision-capable before the user triggers generation
Why Does Open Innovation Matter?
Zero Marginal Cost for Side Hustlers: When starting an Etsy shop or small handmade store, monthly API or SaaS subscriptions add financial anxiety before making the first sale. Open-weight models running locally cost $0 in recurring subscription fees.
Absolute Privacy for Unreleased Products: Artisans and inventors often draft listings weeks before launching novel designs, handcrafted jewellery, or secret recipes. With local open-source inference, proprietary photos and product concepts never leave their hard drive.
Freedom from Arbitrary Vendor Lock-in: Closed AI APIs frequently change pricing tiers, throttle requests, or deprecate models without warning. With open weights, the seller owns their entire software stack permanently.
Niche Customization: Because marketplace template prompts (lib/config/marketplaces.ts) are open code in the repository, any seller can add specialized rules for their niche platform (e.g., Poshmark, Mercari, Depop) without waiting for a SaaS company to build it.
My Agent Session
Prize Categories
eller AI includes a production-ready render.yaml blueprint configured to deploy the Next.js web application frontend on Render. It enables sellers or teams to deploy their studio front end to the web while pointing to their private, secure remote inference server via the OLLAMA_BASE_URL environment variable.
Top comments (0)