Semantic search for Tutorbridge
Meaning-based search across 200,000 documents with filters and permissions. A case study by Rahul Mehta, Machine Learning Engineer (LLMs & RAG).
Challenge
Users of Tutorbridge, a tutoring marketplace, could not find documents unless they remembered exact titles. Keyword search returned too much noise.
Solution
I added hybrid search combining embeddings and BM25, permission-aware filtering, query understanding for dates and authors, and an evaluation harness to tune ranking.