LLM documentation platform · In-house engineer · Fortune 500 manufacturer
An LLM documentation platform: 2 days → 2 hours per document
Engineers at a Fortune 500 manufacturer spent up to two days writing each technical document, and an earlier tool rollout had failed. As the sole engineer, I built a platform that grounds every generation in the source before any model call.
- per document
- 2 days → 2 hours
- referential-integrity score in proof-of-concept testing
- 1.0
- engineers using it
- ~50
The problem
Engineers spent up to two days writing each technical document, and an earlier tool rollout had failed.
What I built
A C# modular monolith that grounds generation in tag-based parsing of the source before multi-agent enhancement on AWS Bedrock.
- Tag-based parsing extracts the facts from the source before any LLM call
- Multi-agent enhancement on AWS Bedrock works from those extracted facts
- One modular monolith, built and run by a single engineer
The result
Documentation went from up to two days to about two hours per document, for about 50 engineers.
In proof-of-concept testing the referential-integrity score was 1.0: zero hallucinated entity names.
Employer work: described without internal names or screenshots.
The service behind this
Production AI Build
Applied AI that holds up in production, not just in a demo.