project · 2025-2026
Schema review assistant
Designed an AI-assisted review workflow that checks proposed event schemas against documented contracts and produces a structured, evidence-backed review.
The problem
Schema and instrumentation reviews require comparing a proposal with contracts, code, and metric definitions. The work is repetitive, yet mistakes are costly because they become downstream data issues.
Approach
- Modelled the review as a workflow of focused retrieval, validation, and explanation steps.
- Grounded responses in approved documentation and versioned schema sources.
- Returned a structured review that distinguishes missing evidence, contract violations, and follow-up questions.
How it works
The replay below uses a fictional event proposal. It shows the useful pattern: retrieve approved context, validate the proposal deterministically, and make the evidence visible to a human reviewer.
Outcome
The design shows how an agent can support review work without replacing the human decision-maker: evidence is surfaced first, and the reviewer retains control of the final outcome.