Product engineer for AI systems. I build AI products end to end: the interfaces, the infrastructure under them, and the verification that makes their output safe to ship.
Currently the sole frontend engineer at an AI financial-modeling startup, where I own the product surface across two stacks (Next.js and an Eclipse Theia/Electron workspace), the contract between the AI engine and the interface, and our Auth0 identity architecture end to end.
My thesis: generation is solved but verification isn't. That's the layer I work in.
If you generate React from Figma with AI, Fixel locks the exact design values and fails CI when they drift. Deterministic token resolution before generation, prohibited-pattern enforcement, spec-locked CI verification, Figma canvas annotation plus an MCP server so coding agents can verify their own output (tested end-to-end with Claude Code).
npx fixel scan ./src # no config, no API key, 10 seconds- Contributions: LangSmith, DeepEval
- Writing: dev.to/amruthakollu — AI verification, design-system enforcement, production stories
- Stack: TypeScript · React/Next.js · Node · Electron · Auth0/OAuth · PostgreSQL/GraphQL · AWS · Claude API · MCP