UPC/product lookup → LLM-crafted search (paid tier) #5

New issue

Open

opened 2026-03-26 22:22:17 -07:00 by pyr0ball · 0 comments

pyr0ball commented

2026-03-26 22:22:17 -07:00

Owner

Summary

Paid feature: user finds a product by UPC, ASIN, or product name in a normalized product database. An LLM then crafts optimal eBay search terms, must-include groups, and must-exclude terms based on known product specs. Results are reviewed and surfaced as they arrive.

Flow

User enters UPC / ASIN / product name
Look up canonical product data (Open Food Facts, UPCitemdb, Amazon Product API, or similar)
Feed product specs to LLM: title, model number, known variants, counterfeit indicators
LLM returns: primary query, must_include groups (model variants), must_exclude terms (box only, empty, broken)
Pre-fill search form with LLM-suggested parameters — user reviews and confirms before searching
Optional: auto-monitor mode — re-run this search on a schedule, surface new listings

Why LLM + human review

Search term quality directly determines trust score accuracy (comps keyed by query hash). LLM knows product variant names, common scam exclusions ("box only", "parts only"), and can generate CNF groups for the must_include mode. Human review before executing prevents wasted searches.

Tier gate

Paid+ (cloud LLM required for quality; self-hosters can BYOK)
Product DB lookup could be free tier if using open databases (UPCitemdb free tier)

Tech notes

Product lookup: UPCitemdb, Open Food Facts (for consumer goods), or barcode scan via vision model
LLM prompt: structured output (query, must_include, must_exclude, category_id)
Pre-fill SearchView form and open save-search dialog with suggested name

## Summary Paid feature: user finds a product by UPC, ASIN, or product name in a normalized product database. An LLM then crafts optimal eBay search terms, must-include groups, and must-exclude terms based on known product specs. Results are reviewed and surfaced as they arrive. ## Flow 1. User enters UPC / ASIN / product name 2. Look up canonical product data (Open Food Facts, UPCitemdb, Amazon Product API, or similar) 3. Feed product specs to LLM: title, model number, known variants, counterfeit indicators 4. LLM returns: primary query, must_include groups (model variants), must_exclude terms (box only, empty, broken) 5. Pre-fill search form with LLM-suggested parameters — user reviews and confirms before searching 6. Optional: auto-monitor mode — re-run this search on a schedule, surface new listings ## Why LLM + human review Search term quality directly determines trust score accuracy (comps keyed by query hash). LLM knows product variant names, common scam exclusions ("box only", "parts only"), and can generate CNF groups for the must_include mode. Human review before executing prevents wasted searches. ## Tier gate - Paid+ (cloud LLM required for quality; self-hosters can BYOK) - Product DB lookup could be free tier if using open databases (UPCitemdb free tier) ## Tech notes - Product lookup: UPCitemdb, Open Food Facts (for consumer goods), or barcode scan via vision model - LLM prompt: structured output (query, must_include, must_exclude, category_id) - Pre-fill SearchView form and open save-search dialog with suggested name

No labels

backlog

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: Circuit-Forge/snipe#5

No description provided.