1A graph-theoretic approach to building reliable LLM judges for retrieval (opens in new tab)(georgianailab.substack.com)3_peregrine_28d ago0Save
2Mixing numeric attributes into text search for better first-stage relevance (opens in new tab)(turbopuffer.com)3_peregrine_2mo ago0Save
3How to build a distributed queue in a single JSON file on object storage (opens in new tab)(turbopuffer.com)2_peregrine_4mo ago0Save
4ANN v3: 200ms p99 query latency over 100B vectors (opens in new tab)(turbopuffer.com)109_peregrine_5mo ago47Save
5Designing inverted indexes in a KV-store on object storage (opens in new tab)(turbopuffer.com)3_peregrine_5mo ago0Save
6Why BM25 queries with more terms can be faster (and other scaling surprises) (opens in new tab)(turbopuffer.com)53_peregrine_5mo ago1Save
7Vectorized MAXSCORE over WAND, especially for long LLM-generated queries (opens in new tab)(turbopuffer.com)2_peregrine_6mo ago0Save
10How we built our own Claude Code (for data) (opens in new tab)(tinybird.co)3_peregrine_10mo ago0Save
14Claude Opus 4 is the best model at SQL generation (opens in new tab)(llm-benchmark.tinybird.live)2_peregrine_1y ago2Save
