Yuvraj Angad S.

I build the layer after the first model call: routing and fallbacks, evals, retrieval, streaming, document pipelines, observability. AI engineer at a health AI product used by millions of people, where I own the LLM routing layer, the memory and retrieval system and document intelligence over medical PDFs. On Upwork I take two kinds of work. AI features that have to hold up in production: LLM integration, RAG, structured outputs, eval harnesses, agent tooling and MCP servers. And precise fixes in large TypeScript and React Native codebases: six Expensify contracts so far, all rated 5.0, every PR reviewed and merged by their maintainers. Open source, with receipts: 20+ PRs merged upstream in GitHub CLI, OpenAI Codex, Gemini CLI, Google ADK, CalCom and Expensify. On the side I build indie dev tools: vibecheck, a linter for AI-generated code with about 2,300 npm installs a month, and brandmd, which turns any site's design system into a design-system file for AI coding tools. How I work: I read the codebase first, find the existing pattern and build on it, and ship the smallest change that fixes the whole problem, with tests. You get a written scope before I start and a PR you can review. Stack: Python, TypeScript, Node, React Native, Next, PostgreSQL, GCP, GitHub Actions. Anthropic, OpenAI and Gemini APIs.