← BackProject · 07
Word Weaver
Bun · Elysia · pgvectorGITHUB ↗
- Built to explore what a writing tool feels like when search understands meaning instead of keywords, finding a note by concept rather than by remembering the exact phrase you used.
- pgvector extension stores OpenAI embeddings for each document chunk. Semantic search queries the vector index using cosine similarity, so a search for 'deadline pressure' surfaces notes about urgency even if neither word appears.
- BullMQ + Redis manages the async embedding pipeline. New and edited documents are queued for vectorization without blocking the write path; the user never waits for an OpenAI round-trip to save.
- Bun + Elysia backend: Bun's native TypeScript execution eliminates the build step in development, and Elysia provides end-to-end typed route definitions so request and response shapes are validated at the framework level.
- Drizzle ORM for schema-first database design: migrations are version-controlled, and schema types flow directly into the application layer without a separate type definition file for database models.
- Docker Compose orchestrates Postgres + pgvector, Redis, and the app server for a fully reproducible local environment. React + Vite frontend with TypeScript throughout.
Tech Stack
BUN
BUNDRIZZLE ORM
DRIZZLE ORMREACT
REACTTYPESCRIPT
TYPESCRIPTDOCKER
DOCKERREDIS
REDIS