Vector search is breaking out of the lab. The buzz covers three paths: Azure Cosmos DB vector search for cloud ease [1], a DIY Comet store built in Go for hands-on control [2], and CoderSwap's DSL-based hybrid search that blends semantic, lexical, and recency signals in real time [3].
Cloud-native vector search — Azure Cosmos DB: This option promises cost-effective, low-latency vector indexing and querying at scale, a sweet spot for teams chasing managed simplicity [1].
DIY vector store — Comet in Go: A from-scratch hybrid vector store that supports BM25, Flat, HNSW, IVF, PQ, and IVFPQ; with quantization, pre-filtering, reranking, reciprocal rank fusion, soft deletes, and index rebuilds. It’s a playground for seeing how internal indexing and ranking work in practice [2].
DSL-based hybrid search — CoderSwap and Oracle: The DSL layer blends semantic, lexical, and recency signals, generating a domain-specific policy at deploy time [3]. At query time the policy fires the first matching intent and streams safe bind parameters into Oracle—no dynamic SQL required.
Which path fits? Cloud for scale and ease, DIY for control, or a DSL-driven hybrid for fast, multi-signal surfacing.
References
Cost-Effective, Low Latency Vector Search with Azure Cosmos DB [pdf]
Discusses cost-efficient, low-latency vector search using Azure Cosmos DB; includes VLDB paper link and methodology.
View sourceShow HN: I built a vector store from scratch (and you can too)
Go-based hybrid vector store; explains indexing, ranking, and persistence; supports BM25, HNSW, IVF, PQ, and quantization.
View sourceBeyond Keywords and Vectors:DSL-Based Hybrid Search
DSL layer blends semantic, lexical, and recency signals for hybrid code search, streaming safe parameters to Oracle in real time.
View source