JevMade field notes / Retrieval evaluation walkthrough
Jevflix
A movie-recommendation walkthrough that separates hybrid retrieval, Jev decisions, gating, fallback, taste-profile behavior, query sets, and retrieval-only versus full-pipeline evaluation.
Original by Ariel BubisRetrievalRepository READMESource reviewed
Before you dive in
What you’ll find in the original
Treat hybrid FAISS/BM25 retrieval as its own stage before asking Jev to make typed decisions over candidates.
Define a gate and fallback explicitly, including how taste-profile mode changes specificity and excludes seed films.
Compare a retrieval-only baseline with the full pipeline on named query sets so constraint handling is not attributed to retrieval.
Worth knowing
Reported evaluation results are repository evidence and were not independently reproduced.