Reading engineering RFCs faster

If your team is doing RFCs right, you've accumulated a hundred. They're the canonical record of every architectural decision — and they're almost impossible to search a year later, when you're trying to remember why you picked Postgres over Mongo.

The problem

RFCs are written by ICs over weeks; they're long, dense, and contain a lot of optional sections (alternatives considered, prior art) that are gold only when you can find them. Notion search and Confluence search are keyword-bound, which is not great for "what did we conclude about transactional outbox?"

How SeekFiles AI helps

• Hybrid retrieval finds the right RFC even when the question doesn't share words with the title.
• Cited answers — "Per RFC-042 §3.2, we chose synchronous outbox for billing because..."
• Cross-RFC Q&A — "have we discussed sharding before, and what were the conclusions?"
• Catches deprecated decisions — "is RFC-018 still in effect, or was it superseded?"

A workflow

1. Build a "Engineering RFCs" assistant. Drop your full RFC archive (export from Notion, Confluence, or Google Docs as PDF).
2. New eng joins → they ask the assistant to learn the architecture instead of pinging seniors.
3. New design needed → check existing thinking before re-litigating: "what have we considered for rate-limiting, and what did we choose?"
4. Architecture review prep → ask the assistant to summarize the active decisions in a domain.

What to upload

• All RFCs (current + archived)
• Architecture decision records (ADRs)
• Postmortems and incident reports
• Major design docs even if they pre-date the RFC process

The point of writing RFCs is institutional memory. SeekFiles AI gives that memory a brain.