The Challenge
Government work has its own cadence. The security bar is higher, the stakeholder list is longer, and "we'll fix it later" does not land well when the client is a regulatory body.
Approach
Started with architecture before features. Set up the component structure, design system, code review process, and development standards first. Then worked through requirements with the government team directly, rather than waiting for a spec to trickle down. Part of that foundation was how the team uses AI: shared rules, custom skills, and review gates so it speeds the work up without giving up the precision a regulated project needs.
Outcome
Still ongoing. The design system is in place, the architecture is documented, and the team has standards they can follow without me in the room.
Key Highlights
- Own the frontend architecture, design system, and code quality across the project
- Set component structure, development standards, and code review norms before writing the first screen
- Work directly with government counterparts and backend engineers to turn requirements into shippable code
- Operating in a regulated environment where security and accessibility are non-negotiable
- Set up the team's AI engineering standards: one rule set that Claude Code, Cursor, and Codex all enforce the same way, kept in version control so conventions cannot drift
- Built a custom AI skill that audits the codebase against the formal business requirements, answering with citations and routing each check by user role
- Put repetitive work (verified PRs, locale scaffolding, content migrations) behind AI commands that pass lint, type checks, tests, and git hooks before a human signs off