# Nityashree S: AI engineer

Hi, I’m Nitya, an AI engineer. I build AI agents that hold up in production.

Most of my days go into LLM agents: the MCP connectors they use, the evals that tell me they work, and the guardrails that keep people safe. I like the quiet part of the job, where I prove something works before anyone has to trust it.

Resume: https://nityashree.in/Nityashree_S_Resume.pdf

## Selected work

### Case 01: An agent went offline. I found the one file that did it.

**Building an AI agent platform.** The engine, studio and gateways behind company-wide AI agents, and how I kept them running.

- Stack: Go, TypeScript/React, PostgreSQL, MongoDB, Docker, Kubernetes, AWS
- Case study: https://nityashree.in/work/agent-platform/ (markdown: https://nityashree.in/work/agent-platform.md)

### Case 02: Agents that see only what you’re allowed to see.

**Building MCP connections.** Connectors that let agents use real apps with each person’s own access, never a shared key.

- Stack: Go, MCP, OAuth 2.0 with PKCE, token vault, structured logs and metrics
- Case study: https://nityashree.in/work/mcp-connections/ (markdown: https://nityashree.in/work/mcp-connections.md)

### Case 03: I attacked our agents before anyone else could.

**Security across an agent fleet.** Every layer closed by default, and every fix proven by replaying the attack.

- Stack: Go, TypeScript, signed webhooks, govulncheck, container sandboxing
- Case study: https://nityashree.in/work/security/ (markdown: https://nityashree.in/work/security.md)

## Try my side projects

Small things I built on my own, each live and open source. The demos run on sample data or simulated model answers, so no sign-up or API key is needed.

- **Smart Router.** An OpenAI-compatible LLM gateway that sends each prompt to a model tier by rules and estimates what it costs. Try it: https://smart-router-nityashree.vercel.app · Code: https://github.com/Nityaa08/smart-router
- **Meridian.** A multi-agent research API: it splits a question, searches each part, ranks the evidence and streams a cited report. Try it: https://meridian-nityashree.vercel.app · Code: https://github.com/Nityaa08/meridian
- **Cross-Border Fee API.** The all-in price of a cross-border payment into Latin America, with every fee and the rule behind it on its own line. Try it: https://cross-border-fee-api.vercel.app · Code: https://github.com/Nityaa08/cross-border-fee-api
- **Fraud Pattern Explorer.** A risk dashboard where every chart is also a filter, to get from a suspicious spike to the transactions behind it. Try it: https://fraud-pattern-explorer.vercel.app · Code: https://github.com/Nityaa08/fraud-pattern-explorer

## How I work

One laptop, a team of agents, and me in the loop. I don’t hand-write code. I direct AI coding agents, and I prove every change before it ships. Subagents research, challenge and review in parallel, while I set the bar, read every line and own what goes out.

I like working close to the people who use what I build: real questions from Sales and Business Development shaped how I tested, and finance’s own rules shaped the agents I built for them.

1. Plan
2. Delegate
3. Prove
4. Own

Tools I work with: Claude Code and Claude Design every day, plus Cursor, DeepSeek and Grok. Personal agents in Slack run my pull-request workflow. My side projects, an LLM gateway and a multi-agent research API, are on my GitHub.

## My path

On the page this is a hand-drawn trail, a journey with four stops, oldest first: a mortarboard resting on an open book, a payment card in front of a small bar chart, a magnifying glass with a tick checking a chat bubble, and an agent network (a hub with an AI sparkle linked to three tools: a document, a terminal and a shield) with a paper plane landing beside it.

1. **Studied.** Computer science, with a focus on AI and data science, at SASTRA University.
2. **Payments data.** I looked after the payments data an AI assistant answered from, and learnt how much an answer depends on its data.
3. **Testing AI.** I tested that assistant before launch and caught its wrong answers. That’s when AI engineering clicked for me.
4. **Shipping agents.** Then I moved to building and securing AI agents that people rely on at work: the platform, the tools and the guardrails.

## When I’m not working

*Off the clock.* Away from the screen: books, classical dance, yoga, a stage and a mic. Everything on my desk comes with a note.

- **Books.** Any book adaptation out there? I’ve probably read the book first.
- **Ghungroo.** Dance came first, before almost anything else I learnt. Not one but two Indian classical forms, all the way to competition stages.
- **Yoga mat.** Yoga began as a daily ritual and grew into competitions. It taught me that stillness takes practice.
- **Mug.** Not every object needs a story. This one just keeps the desk company.
- **Globe.** Work took me from India to Mexico. Working with people from all over the world was the most fun kind of learning.
- **Run-sheet.** Clipboard in hand, I ran fun and management events for students in college. The best way to bond, and we had a blast.
- **Mic.** Now and then the mic found me: I emceed a few events on my college club’s main stage, and it was a lot of fun.

## Contact

An idea, a question about agents, or simply hello: I’d love to hear from you.

- Email: nityashreesundararajan@gmail.com
- LinkedIn: https://www.linkedin.com/in/nityashree-s-b34b1b291
- GitHub: https://github.com/Nityaa08
- Resume (PDF): https://nityashree.in/Nityashree_S_Resume.pdf