AI and full-stack developer · 3rd-year Computer Science (CSE) student in Bengaluru
I build AI systems end to end: agents, RAG and AI security. I don't trust a feature until I've measured it, so every number below comes from a test or an evaluation you can re-run from the repository.
Portfolio · Resume (PDF) · LinkedIn · Email
| Project | What it is | Measured |
|---|---|---|
| Dwarpal · live demo | Security gateway between AI agents and their MCP tools. Stops tool poisoning, tools that change after approval, and private data leaking to the wrong server, in English, Hindi, Hinglish and Kannada. | Data leaked in 0 of 25 attack runs with Dwarpal vs 5 of 25 without, 0 false blocks; Helm chart tested on Kubernetes in CI |
| AyurIP-Sahayak · live demo | Multilingual RAG assistant for Ayurveda and Indian IP law that cites the exact source page for every claim. | 0.99 answer faithfulness (166 of 167 claims), Hit@1 0.95, 47 tests |
| NestFinder · live demo | Tool-calling AI agent that finds flats in Bengaluru, explains match scores and checks fair rent. | Tool-choice accuracy 20/20 on its eval, every model call traced in Langfuse |
- AI / LLMs: RAG, AI agents, LangGraph, Model Context Protocol (MCP), tool calling, embeddings, vector search, LLM evaluation (LLM-as-a-judge), guardrails and prompt-injection defence
- Backend: Python, FastAPI, Node.js, Express, PostgreSQL + pgvector, Redis, MongoDB
- Frontend: React, Vite, Tailwind CSS
- DevOps: Docker, Kubernetes, Helm, GitHub Actions, Vercel, Render
Founding Software Engineer Intern, FERMOR (Sep – Oct 2026): redesigned the website and its SEO; launch day brought about 30x the usual daily traffic.
Learning to fine-tune small models (QLoRA) to replace Dwarpal's policy judge. Open to AI/ML and full-stack internships.