harsh-kumar.dev › aboutscore 9.91
Final-year CS student building search and retrieval systems — and, more importantly, building the evaluation harnesses that prove whether a change made them better or just different. Backend by training, retrieval by obsession. Working rule: measure, don't guess.
github.com › sgnhyperion › AI_Search_Enginescore 9.87
Search over 120,000 documents combining a hand-built BM25 inverted index, FAISS vector search, and cross-encoder reranking, served over FastAPI. Currently being rebuilt from a static-dataset demo into a real engine: SQLite-backed incremental indexing, chunk-level retrieval, and a polite web crawler.
pythonbm25faisssqlitefastapi
harsh-kumar.dev › writingscore 9.42
No posts indexed yet. The first one lands August 2026: building an evaluation harness for a hybrid search engine, and what nDCG@10 said about fusion weights everyone just googles.