Back to projects04 / Procurement / product intelligence
AI Product Sourcing Agent
The AI Product Sourcing Agent is a procurement intelligence layer that compares marketplace offers on consistent evidence instead of treating search results as interchangeable products.
The problem
Cross-marketplace sourcing creates a comparison problem: listings use different currencies, units, terminology, supplier evidence and duplicated offers, so search volume alone does not produce a procurement decision.
How I approached it
- 01Plan multiple search queries instead of betting the sourcing result on one wording.
- 02Normalise offers into a shared product/supplier model.
- 03Deduplicate repeated or near-identical listings before ranking.
- 04Rank with observed evidence and preserve provenance for the operator.
03 / WHAT SHIPPED
Key features
- Query planning
- Marketplace abstraction
- Offer normalisation
- Cross-market deduplication
- Evidence-based ranking
- Supplier/product provenance
Architecture
- Core
- TypeScript
- Search
- Multi-query planning
- Normalisation
- Shared offer model
- Ranking
- Observed evidence
- Output
- Comparable candidates
- License
- MIT
What SOURCING is built to preserve.
- Evidence before enthusiasm
- Comparable units before ranking
- Provenance beside every candidate
- No fake supplier certainty
Useful answers, without the hunt.
Is the sourcing agent tied to one marketplace?
No. The public engine is marketplace-agnostic and is built around normalising evidence from different sources into a comparable model.
How does it reduce duplicate results?
It includes cross-market deduplication so repeated or near-identical offers do not dominate the ranked candidate list.
Does it guarantee supplier quality?
No. It ranks candidates from observed evidence and provenance; supplier verification remains a separate procurement step.
