Databases / prototype

Vector DB

A multimodal vector database engine in Rust using RocksDB for storage, custom embedding modules, KD-Tree indexing, and KNN search.

Architecture Overview

Vector DB stores embeddings and metadata in RocksDB, then exposes vector search through custom indexing and retrieval components.

The project explores a compact database architecture: persistent storage, embedding generation hooks, indexing structures, and terminal-first interaction.

Technical Challenges

storage layout for vectors and metadata
KD-Tree construction and query performance
KNN search implementation tradeoffs
module boundaries for custom embedding generation
terminal UI ergonomics for inspecting database state

Benchmarks

The key measurements are index build time, query latency, recall quality, and RocksDB storage overhead as vector counts grow.

Lessons Learned

Vector search systems are database systems first: storage layout, indexing strategy, and observability matter as much as the embedding model.

Tools

Rust, Ratatui, RocksDB