Overview
A product-discovery platform for B2B industrial buyers (RFQ-first, no cart or checkout) that resolves a plain-language problem into a structured spec, ranks matching products and turns the shortlist into a Request for Quotation. It runs with no database or API key by default.
Highlights
- Six-stage engine: intent parse, hard constraints, BM25 fused with vector cosine by reciprocal rank, per-category scoring, brand-diversified selection, generated explanation.
- The LLM only converts language into a schema and rephrases computed results; every number comes from a unit-testable scoring model.
- 241 Vitest tests, 154 in-process smoke checks and 96 HTTP smoke checks; 84 pages prerender.