Jaideep Ray
Sr. Staff ML Engineer. Author of writing on multi-stage ranking in search systems — the architecture of separating a fast, high-recall retrieval stage (BM25, ANN) from a slower, high-precision reranking stage (cross-encoders, LLM-based scoring).
Key Contributions
- Multi-Stage Ranking: Articulated the engineering tradeoffs in two-stage pipelines — why retrieve-then-rerank outperforms either stage alone, how to set the retrieval recall budget (top-k candidates passed to reranker), and the latency/quality Pareto curve for each stage.