RefundAI
AI refund agent with policy enforced in deterministic code
An AI agent that approves, denies, or escalates refunds against a strict policy — with every hard rule enforced in deterministic code, not the LLM, as the single source of truth.
Software Engineer — backend, GenAI & real-time voice.
I build production backend, AI, and real-time systems that solve real business problems.

I'm a Software Engineer with 1.5 years at Think41, where I built and shipped production systems across three products — Fabric (an internal multi-tenant GenAI workspace used across the org), Entero (a B2B/B2C healthcare platform), and a real-time voice AI agent.
My focus is backend and GenAI: designing APIs and data models, wiring LLMs and agentic workflows through MCP, and keeping critical decisions in deterministic code rather than the model alone — so a system is observable, safe, and correct.
Beyond work I hold an Indian patent and an IEEE publication on real-time driver-alertness monitoring, and I'm Oracle Cloud GenAI Certified.
Python, FastAPI, Django, Node.js. REST, WebSockets, and Server-Sent Events for real-time streaming.
LLM integration, agentic workflows, MCP servers, RAG, and prompt-injection defense.
Pipecat, Deepgram, ElevenLabs on AWS — streaming pipelines tuned for low latency.
AWS (EC2, S3), Docker, Linux, Git, CI/CD.
Production systems and side projects across backend, GenAI, and real-time. Each has a case study.
AI refund agent with policy enforced in deterministic code
An AI agent that approves, denies, or escalates refunds against a strict policy — with every hard rule enforced in deterministic code, not the LLM, as the single source of truth.

Internal multi-tenant GenAI workspace used by 70+ employees
A multi-tenant GenAI workspace that lets teams run LLM workflows over Gmail, Calendar, Drive, HubSpot, and WhatsApp — with agent steps streamed live for transparency.
Streaming voice pipeline, 40% lower end-to-end latency
A production real-time voice agent built on Pipecat, Deepgram, and ElevenLabs — re-engineered to cut end-to-end latency by 40% (5s → 3s).

B2B/B2C healthcare commerce platform
Core backend and mobile features for a healthcare commerce platform — lab-test bookings, medicine orders, and 10+ production modules.

LLM-powered inventory dashboard
An AI inventory dashboard that turns uploaded sales data into analytics, demand forecasts, and restock recommendations — built solo.
WebSocket game-state sync, chat & live leaderboard
A full-stack real-time multiplayer game with WebSocket state synchronization, in-match chat, and a live leaderboard — deployed to the cloud.
1.5 years shipping production systems at Think41, across three products.
Jul 2025 – Jun 2026 · Bengaluru, India
Feb 2025 – Jun 2025 · Bengaluru, India
Sep 2023 – Nov 2023 · Remote
B.Tech, Computer Science and Engineering · Kattankulathur, Chennai
2021 – 2025
CGPA 9.21 / 10
The stack I reach for across backend, GenAI, and full-stack work.
A patent, an IEEE publication, and certifications backing the work.
Patent — IP India
Application No. 202441068474
IEEE — ICESIC 2024
Peer-reviewed, accepted in IEEE Proceedings.
Hackathon
Real-time weather-report mobile app.
Oracle
Open to opportunities · Relocation friendly — based in Bengaluru, India. The fastest way to reach me is email.