Challenge and role
The challenge
AI tooling in a bank has to be reusable and accountable, not a drawer of one-off demos.
The responsibility
I set the technical direction and lead delivery across the MCP layer and the evaluation that gates it.
Decisions
- 01
Turn recurring engineering problems into tools other teams can use.
- 02
Design MCP servers and schemas that carry approved design intent between tools.
- 03
Build evaluation and human review in before anyone adopts it.
Outcomes
Moved AI work from side-of-desk experiments to shipped internal tooling.
Built MCP servers reaching Figma, GitHub, Atlassian, Confluence, Ardoq, Backstage and Harness.
Built visual diffing and human approval into the pipeline rather than beside it.