November 2025
Vector Embeddings
An experiment in building a semantic search pipeline from scratch: text goes through OpenAI's embeddings API, gets stored in SQLite via sqlite-vec, and is queried by cosine similarity for the closest matching documents.
Semantic search pipeline built from first principles.
Semantic search pipeline built from first principles.
An experiment in building a semantic search pipeline from scratch: text goes through OpenAI's embeddings API, gets stored in SQLite via sqlite-vec, and is queried by cosine similarity for the closest matching documents.
More projects
view all →SQLite Vector Store
A semantic search system for knowledge graphs using SQLite with the sqlite-vec extension and OpenAI embeddings, so knowledge-graph JSON files in object storage can be found by meaning, not just keyword matching.
Chara Gen DDD
A small DDD proof of concept for AI character-image generation — generates a character in different art styles from one base reference image, deliberately scoped to the domain layer only (no CLI/UI layer) to test the modeling approach first.
MCP POC
A proof of concept for AI-driven workflow orchestration — an LLM generates a YAML (or JSON) workflow spec, executed against a fixed set of predefined TypeScript functions, so building a new workflow becomes a generation problem rather than new code every time.