hackathon / prototype
Parichay
AI-powered material-code harmonization for public-sector enterprises.
Built a material-code harmonization prototype for CPSEs on FastAPI and PostgreSQL/pgvector, combining 384-dimensional sentence embeddings, FAISS cosine search, BM25 lexical retrieval, and NetworkX cluster validation.
Implemented critical-attribute safety checks to block unsafe material matches.
Synthetic benchmark: 99.66% Recall@30 with zero unsafe merges. The benchmark described in my résumé used 38 canonical materials, 168 variants, and 14 hard-negative groups. These are prototype/synthetic benchmark results, not production deployment metrics.
Context: Smart India Hackathon 2026.