2. Building A Roadmap

Building a Roadmap

Next Steps:

Roadmap proposed by Claude:

Step	Title	Description
Step 1	Upgrading Our Datastore	Swap index.pkl for pgvector Get a real vector database in place first. Everything else gets easier once you have proper persistent storage with metadata filtering. This is also where your Spring Boot background becomes relevant — pgvector is just Postgres
Step 2	Adding Hybrid Search	Once you’re on pgvector you can add keyword search alongside vector search almost for free — Postgres has full text search built in. This immediately improves retrieval quality before you add more data
Step 3	Adding more data sources	Now that the foundation is solid, add the codebase first — it’s simpler than Slack because it’s just files. Slack is the messiest source — short messages, lots of noise, threading makes chunking tricky. Save it for last
Step 4	Spring Boot API layer	Wrap everything in an API so it’s not just scripts anymore
Step 5	Slack integration	Now you have something worth wiring up to Slack

Today’s work: Step 1 has been completed - we have succesfully migrated over from a index.pkl file -> using a pgvector database which our RAG application succesfully calls to retrieve context. our previous process of loading the index file, then finding the relevant chunks is now streamlined into 1 function where we embed the query and directly use cosine similarity to find the the top_k relevant contexts.

To do: we’ll have to figure out how to avoid storing duplicate context - because everyime we run ingest_confluence.py, we will just add the same information again