A production LLM pipeline that finds what slows engineers down
Development transcripts, scrubbed of secrets and personal data, read by Claude into typed findings. The confident ones arrive as pre-filled tickets.
A tool that reads engineers' transcripts has two ways to fail. It can send something it should never have sent, or it can produce findings nobody acts on. The second one kills it quietly.
- Anthropic API
- Claude
- structured output
- Go
- Node.js
- OAuth 2.0 device flow
- Slack
- Jira
