This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
What I Built
Solo artisans, woodworkers, and vintage sellers face an exhausting bottleneck when expanding an e-commerce side hustle: listing friction.
Handmade merchants spend upwards of 30 to 45 minutes per product manually researching high-volume keywords, formatting narrative descriptions, and strictly staying within Etsy’s rigid search criteria:
- Titles capped at 140 characters (where early keyword density determines click-through rate)
- Exactly 13 algorithm tags, each capped strictly at 20 characters
- Conversion-oriented bullet points outlining materials and care instructions
ShopCraft AI solves this problem in under 5 seconds. A seller simply enters raw notes—the item name, materials used, target audience, and key highlights—and ShopCraft formats a marketplace-ready, high-converting product listing adhering strictly to e-commerce SEO requirements.
Who Would Pay for It?
Solo creators on Etsy, Shopify, and Depop. Small makers cannot justify $20–$50/month enterprise SEO platforms like Marmalead or eRank when they only launch 5–15 products a month. A hyper-focused micro-SaaS priced at $5/month or $0.20 per batch listing offers an immediate return on investment for small shop owners.
Demo
- Live Deployed App: https://shopcraft-ai-htvrhihrounmvh2qzkvsyj.streamlit.app
Try generating a listing with the default sample:
- Product: Wheel-thrown Stoneware Matcha Bowl
- Materials: White stoneware clay, food-safe moss green reactive glaze
- Target Audience: Ceremonial matcha drinkers, minimalist home aesthetic
- Features: 12oz capacity, built-in spout for pouring, microwave & dishwasher safe
Code
You can view the full repository, requirements, and deployment setup here:
How I Built It
ShopCraft AI is built around Google's open-weight model, Gemma 2 (2B Instruct).
- Frontend & Orchestration: Built using Python and Streamlit, providing an uncluttered, responsive seller workspace.
- Model Layer: The app integrates with Google Gemma 2 2B-IT, utilizing prompt instruction tuning to enforce structured field generation (Title, 13 SEO Tags, and Formatted Description).
- Format Constraint: Because marketplace algorithms reject tags exceeding 20 characters, the prompt explicitly constrains Gemma 2 to produce comma-separated phrases under this character ceiling, avoiding post-processing failure.
- Deployment: Hosted directly on Streamlit Community Cloud with continuous integration linked to GitHub.
Why Does Open Innovation Matter?
Why build this on an open-weight model like Gemma instead of relying on closed commercial APIs like OpenAI or Anthropic?
1. Sustainable Micro-SaaS Unit Economics
Closed APIs charge per prompt token and generated token. For an indie side hustle targeting micro-merchants at $5/month, unpredictable token usage and bursty scraping rapidly destroy profit margins. Gemma 2's open weights allow self-hosting, local deployment, or predictable fixed-cost container deployment where inference costs stay stable regardless of request volume.
2. Protecting Proprietary Seller IP
Handmade artisans frequently input confidential formulas, unreleased collections, and novel design techniques. With open models running on controlled instances, customer design notes and proprietary material compositions are never routed into closed third-party datasets for model retraining.
3. Model Independence & Longevity
Closed API providers routinely deprecate endpoints, alter hidden system prompts, or modify safety thresholds that break structured JSON/formatting parsers without notice. Open-weight models like Gemma 2 give builders permanent architectural stability: once the prompt and weights produce the intended schema, the service remains consistent indefinitely.
My First User
To test ShopCraft AI in the real world, I shared the live deployment with a local creator who sells handmade terracotta crafts and customized home decor.
The Test: She entered a listing for a hand-painted terracotta desk planter.
Her Feedback:
"Usually writing the tags is the most annoying part because Etsy cuts you off at 20 characters, and I keep having to re-count letters. The app gave me 13 tags that were ready to copy-paste. I'd definitely pay a small flat fee like ₹200–₹400 ($3–$5) a month if it saves me an hour every weekend."
Next Request:
She asked for an export button to copy all 13 tags into clipboard memory with a single click, which is slated for the next version update.
Prize Categories
- Best Use of Gemma
Top comments (0)