IRDB - Learning Hybrid Search in PostgreSQL & Rust

A study project exploring hybrid search by combining BM25 full-text search with vector similarity search in PostgreSQL.

What is This?

This project demonstrates how to combine two complementary search approaches:

• Lexical search (BM25) - Keyword matching for brands, models, technical terms
• Semantic search (Vector) - Meaning-based similarity for natural language queries
• Weighted combination - Blending both methods (30% BM25 + 70% Vector)

Key Technologies

• PostgreSQL 17.5 - Latest stable PostgreSQL
• ParadeDB pg_search 0.20.2 - BM25 full-text search with custom operators
• pgvector 0.8.0 - Vector similarity search with HNSW indexing
• Rust - Type-safe, data-oriented API layer
• Leptos 0.7+ - Reactive web framework (in progress)
• CloudNativePG - Kubernetes operator for high availability

Documentation

Comprehensive documentation is available in the docs/ directory:

1. Architecture Overview - System design, technologies, and architectural decisions
2. Deployment Guide - Docker Compose and Kubernetes deployment
3. Hybrid Search Deep Dive - Implementation details, algorithm, and examples
4. Web Application Development - Building the Leptos-based UI
5. References & Resources - Papers, documentation, and learning resources
6. Testing Guide - Testing and coverage notes

License

Apache License 2.0 - See LICENSE file for details.

This is a study project demonstrating hybrid search techniques. The code is provided as-is for learning and research purposes.

Acknowledgments

Built with open-source technologies:

• PostgreSQL Global Development Group - Core database
• ParadeDB Team - BM25 search extension
• pgvector Contributors - Vector similarity search
• Leptos Team - Rust web framework
• CloudNativePG Team - Kubernetes operator
• Rust Community - Language and ecosystem

---

Study Focus: Hybrid Search in PostgreSQL & Rust Last Updated: 2025-12-17

© 2025 sojoner