VVashni Roy·2d agoDoes a Higher Similarity Score Mean Better RAG Retrieval?I tested small, medium, and large chunks in RAG and found that the highest similarity score didn't always give enough context to answer the question.4 min read·vashniroy.hashnode.dev
RRisingWave Labs·Nov 9Tutorial: A Customer-Support Agent with Live Context in 120 Lines of SQLKafka, Postgres CDC, in-stream embeddings and agent memory as materialized views on RisingWave9 min read·risingwave.com
SSai Kiran Kshatriya·Nov 5Recall vs Latency: Tuning CockroachDB Vector Indexes in ProductionThree controls, one paired metric, and a benchmark method for finding the bend in the recall–latency curve.15 min read·saikirankshatriya.hashnode.dev3A
SSai Kiran Kshatriya·Nov 5Bringing AI Context Closer to Users with CockroachDB Table LocalitiesOne logical table, each tenant's AI memory homed in their region with REGIONAL BY ROW.18 min read·saikirankshatriya.hashnode.dev2SA
AAritra Chakraborty·Oct 26Databricks Lakehouse Platform for Real-Time, Agentic Data and AI WorkloadsFederation, Lakeflow, Unity Catalog, Lakebase, AI Search and agents on one governed platform, with real-time patterns and limits.4 min read·aidatalegends.hashnode.dev
AAritra Chakraborty·Oct 26Databricks Runtime Introduces NEAREST BY: Native SQL Vector Search at ScaleBatch vector search in SQL with a fused Photon GEMM operator and IVF indexes in Delta tables: how it works and when to use it.4 min read·aidatalegends.hashnode.dev
MMuhammad Tahir·Oct 11Hybrid Search Architecture: Combining BM25 Keyword Search with Dense Vector EmbeddingsThis article was originally published on Muhammad Tahir's Portfolio. Introduction & Industry Context In the early wave of generative AI and Retrieval-Augmented Generation (RAG) deployments, engineerin13 min read·mtdeveloper.hashnode.dev
Mmarket·Sep 28Correct where= cut 12 cross-product bleeds — meta_purity 0.65→1.0 on our 96-chunk Chroma setSame author multi-product corpus (96 chunks, products A/B/C + locale/version) on Chroma where=. Compared vector-only vs correct $and vs wrong product. Desk API: $0. pipeline P@5 Recall@5 empty% met1 min read·eworldlive.hashnode.dev
MMohammed Yahya Ahmed·Sep 24From Embeddings to Search Architecture: Building BrainZen’s Retrieval FoundationLearn how BrainZen turns content into embeddings and uses hybrid search, BM25, RRF, and structured filters to build a hybrid search system.7 min read·learningwithyahya.hashnode.dev1
MMohammed Yahya Ahmed·Sep 22Choosing the Embedding Architecture for BrainZenHow I chose browser-side embeddings for BrainZen, comparing API, backend, and inference-server architectures and their tradeoffs.5 min read·learningwithyahya.hashnode.dev11A