Microsoft.SemanticKernel.Connectors.SqliteVecでサーチが動かん、と1日悩んだが、バージョンのせいだった…。 SqliteVec→1.74-prev →(depend)Microsoft.Extensions.VectorData.Abstractions 10.1.0 なんだが、VectorData.Abstractionsの最新版10.7でSqliteVecが必要にしている関数が消えている…

SQLiteを利用したハイブリッド検索対応の軽量RAGの実現 acro-engineer.hatenablog.com… #生成AI #RAG #SQLite #AWS #Bedrock #FTS5 #sqlitevec #テックブログ

RAG、ベクトル検索だけだと 日本語の固有名詞・略称・ID検索で取りこぼすこと、ありませんか？ 今回の記事では、SQLite sqlite-vec FTS5 で 全文検索とベクトル検索を組み合わせた “軽量ハイブリッド検索RAG”を実装。 段階式検索やBM25リランクも比較し、 軽量構成でも高精度な検索を目指す方法を紹介しています👇 acro-engineer.hatenablog.com… #生成AI #RAG #SQLite #AWS #Bedrock #FTS5 #sqlitevec #テックブログ

New op-sqlite release. Speeds up the loading of dylibs (sqlitevec, crsqlite, libsql) quite a lot (which also speeds up the app boot speed github.com/OP-Engineering/op…

new op-sqlite release - Minor perf improvement on Android - Big speed up on iOS start up if not using crsqlite/sqlitevec by skipping bundle lookups github.com/OP-Engineering/op…

We are excited to have @TiemmaBakare as a speaker for #DevFestLagos25. Emmanuel Bakare is a Senior DevOps Engineer with experience spanning Systems, Networking, and FinOps. In this talk, he will take you through an exciting journey on the various design choices engineers can consider with databases, preferential paths in designing cost-effective solutions with paths like SqLiteVec or enterprise vector solutions like Convex, S3 vectors, and many more. Get your ticket at devfestlagos.com

I voted LangChain because it's the only one I've used. The documentation is great, tons of out of the box packages and integrations, lots of good examples, seems to be the most common, and LangGraph is also very nice. LlamaIndex seems focused on RAG. For my job, I built a SQLiteVec RAG app using LangChain and their SQLiteVec vectorstore class (which I modified to allow for metadata pre-filtering). I also used LangGraph to build a workflow for the Chat interface on Staqc.com, and it's really nice that they have a draw method which outputs a mermaid diagram of your workflow/agent system.

I'm creating a vector store database using SQLiteVec for each document and then record for the document in my primary database for each document. That was the simplest way for me upload, filter documents, and delete documents (and their respective vector stores). I can add attributes to the document in the primary database for filtering purposes, then use similarity searching for retrieving documents and giving them to the LLM for RAG.

sqlite-vec support is out! Just add sqliteVec: true in the package.json and you can start using embeddings in your AI app github.com/OP-Engineering/op…