Vector Search vs Keyword Search vs LLM SQL: Which is Best for SQLite?
Comparing three approaches to searching a SQLite database: vector similarity search (sqlite-vec), keyword search (FTS5), and LLM-generated SQL queries.
In this video we explore: • Setting up sqlite-vec for semantic vector search • Using sentence-transformers for embeddings • Generating SQL queries with Claude • Trade-offs: speed, accuracy, cost, and explainability • When to use each approach (and when to combine them)
Using a real-world dataset: 28k+ synthetic customer support tickets from Hugging Face.
Libraries covered: sqlite-vec, sentence-transformers, polars, lisette
🔗 Resources:
sqlite-vec: https://github.com/asg017/sqlite-vec
Dataset: Tobi-Bueck/customer-support-tickets on Hugging Face
Tags: sqlite, vector search, embeddings, FTS5, LLM, SQL, python, sentence-transformers, sqlite-vec, semantic search, RAG, retrieval, machine learning, Claude, customer support, polars, hybrid search
https://buymeacoffee.com/alexkelly