Inspiration Product recalls cost billions and often trace back to selling somewhere you shouldn't have. We wanted a tool that answers "Where can I sell this?" in minutes, not months.
What it does Paste your website URL or upload a product label and get a color-coded world map in under 60 seconds — green (sell freely), yellow (review needed), red (blocked) — across 20+ countries, all 50 US states, and major retailers like Whole Foods and Walmart. No account required.
How we built it Three-phase pipeline on Next.js 16: (1) URL scraping or file upload with OCR/PDF parsing extracts product data; (2) parallel Tavily searches pull live regulatory evidence per region; (3) Pioneer GLiNER2 compresses raw regulatory text into structured entities (~60% fewer tokens) before parallel LLM calls via Pioneer's OpenAI-compatible API produce per-country, per-retailer, and per-state verdicts. Results are backed by a ClickHouse FDA recall corpus and Supabase for user data.
Challenges we ran into Getting the LLM to return specific, differentiated verdicts per country — early prompts defaulted everything to "review." Strict prompt engineering and parallel region batches fixed it. We also built a custom JSON salvager to handle truncated model responses without crashing the full report.
Accomplishments that we're proud of The Pioneer GLiNER2 pre-pass cuts a 4 kB regulatory article down to ~1.6 kB of structured entities before it touches the LLM — real cost and latency savings. And every thumbs-up/down feeds Pioneer's adaptive retraining loop, so the extractor improves permanently at zero cost per retrain.
What we learned Live web evidence + a compact NER pre-pass beats stuffing raw documents into a large context window. Small, focused parallel LLM calls are faster and more reliable than one giant prompt.
What's next for Recall0 Automated alerts when regulations affecting your saved products change, deeper ingredient analysis using FDA recall history, and shareable compliance report PDFs for supplier audits.
Log in or sign up for Devpost to join the conversation.