SupplierVerse.ai
An AI procurement platform that turns fragmented public tenders into grounded, reviewable bid work across discovery, matching, collaboration, and response drafting.
The operating problem
Procurement teams needed to find relevant tenders across seven inconsistent public portals, coordinate opportunity data, and draft answers without letting a model invent evidence.
My contribution
Worked across the Next.js product, 85 FastAPI endpoints, 297 database migrations, ingestion jobs, matching, collaboration, billing, and shared identity boundaries. Built the Azure OpenAI and LangGraph Bid Writer flow around typed state and reviewable output.
System shape
The key decision
The Bid Writer uses a six-node graph with evidence allow-lists, claim provenance, critique loops, missing-input handling, and human insertion instead of autonomous submission.
Reliability by design
Capture, normalisation, publication, enrichment, matching, explanation, feeds, and notifications are separate recoverable stages. SSE exposes progress without hiding failure.
What the source proves
Honest limits
The current audit does not establish broad production traffic, matching accuracy, or unattended scheduling for every source.