Podacle

At Podia, a small team works on a 10+ year old codebase, and no single person holds complete knowledge of how every feature interacts. Changes in one area can cascade unexpectedly into another — deleting a product cancelling Stripe subscriptions, soft-deletion tripping unique constraints.

Podacle (“Podia oracle”) is an experiment in capturing that institutional knowledge as a queryable graph. It parses the Rails codebase to extract model associations and callbacks, service object dependencies, controller actions, background job chains, and external service calls, then stores the relationships as RDF triples with confidence scores and source tracking.

On top of the graph sit a SPARQL query engine and an MCP server, so you can ask questions in natural language — “what happens if I delete a product?” — and get answers grounded in the actual code relationships.

How it was built

Ruby end to end: a knowledge extractor walking the codebase, an RDF triple store, and an MCP server exposing the query engine to LLMs. It’s a weekend-scale experiment rather than a production tool, but it’s a direction I keep coming back to: the graph doesn’t hallucinate, and the LLM doesn’t have to hold the whole codebase in context.