use LLM embeddings to seed information at varying levels of hierarchy for learning improved embeddings over the codebase
- AST for types
- keywords, functions classes, variables, etc.
- openai embeddings over sliding windows of tokens
- learn embeddings for respective functions, classes, lines of text, etc