Engineered a full-stack RAG chatbot for UP Diliman, currently in beta helping 200+ CS students and advisers with enrollment and university-related concerns.
Designed a multi-stage retrieval pipeline: query classification, rewriting, multi-query expansion, hybrid search (cosine + BM25 in PostgreSQL with pgvector), cross-encoder reranking, and citation-mapped streaming generation.
Built a RAGAS-based evaluation framework with 97% faithfulness, 95% answer relevancy, and 85% factual correctness based on initial testing.