Report percentiles, not just the mean.

Search evals are broken. We track p50/p99 for latency because an average hides the tail, then report search quality as one average NDCG and call it a day. A mean of 52.5 can sit on top of a fifth of queries finding nothing at all. Track p25 and p50 alongside the mean.

Stop Averaging Search Evals

Full Quality, 13× Faster Queries

Inference-free SPLADE nearly matches full SPLADE quality at 13× lower query latency, no GPU needed at query time. Benchmarked against full SPLADE and BM25 on Qdrant and pyserini.

Inference-Free SPLADE

Search is everywhere and is one of the hardest problems in CS. Here's why I'm obsessed with it.

Why I'm Obsessed With Search

I Reverse-Engineered Exa.ai Infrastructure Cost with Napkin Math

Exploring the JSON embeddings to for matching new products with existing ones