EdgeDB is the knowledge graph of their RAG “EdgeDB as the knowledge graph of Dexa” Did 150,000 queries a day CPU went to 1.25 15,000 queries in one hour 0.1 CPU and less than a gig of memory
- Strong opening describing Dexa
- Mention Huberman partnership
- Stack: TypeScript (why they started with EdgeDB)
- 100% type safe from data ingestion to backend
- Type checking (via TS strict mode) is a good indicator everything is working
- Takes the place of testing
- Data ingestion is very separated from the app; EdgeDB helps maintain the separation
- EdgeDB is a good way to tie together a monorepo
- Long-term mission: become a much better search engine
- Serve results to people you trust with direct attribution to audio, video, and text results
- They will publish on their blog
- Codebase: 26 packages in the monorepo
- half interface with the DB
- EDB schema and types get used in some form in most of them
- Can stay in sync only because of the EDB types being used across the monorepo
- Riley will share schema
- They use a lot of computed properties
- roll up a computed property
- UI usage: “It’s pretty good.” Riley spends more time in the CLI REPL
- Uses schema viewer
- REPL feedback
Images
Transcript
- 8:40- architecture