AI Systems & Backend Engineer · Bengaluru
My main tech stack is Python and TypeScript , with FastAPI and Next.js for building scalable full-stack applications .
I build AI Agents systems using LangChain and LangGraph , and deploy real-time voice agents with Pipecat .
For infrastructure, I use AWS and Docker , and manage data with Neo4j , PostgreSQL , and Redis .
Founding Engineer
Dec 2025 — Present · Hybrid
Building Zaum's Harness as a Service (HaaS) layer: the agent harness that makes AI outcomes reliable and the underlying model swappable, because the work on the harness compounds.
Shipped multiple 0-to-1 products on the road to Zaum: an AI investor-simulation platform (meetchamp.in/ai) that acquired 200 users on launch day, and an agent-to-agent networking product where professionals deploy AI digital twins.
Engineered a "second brain" system from scratch using context graphs, autonomous learning modules, and dynamic persona creation, deployed and self-hosted on our own infrastructure.
One of three engineers owning backend architecture, AWS infrastructure (ALB, SQS), and production deployments end to end.
A look back at my work. Selected archives showcasing projects across AI, voice, and full-stack.
Autonomous A2A Networking Platform
Building an Agent-to-Agent (A2A) network where real professionals can create their AI digital twin in under 2 minutes to autonomously network and evaluate synergies on their behalf. Integrated seamless Telegram connectivity for agent interaction.
Universal Smart Device Enabler
Engineered a custom hardware-software bridge using ESP32 microcontrollers that can retrofit any standard physical appliance into an intelligent, AI-controlled IoT device. Near-instant execution (~530ms latency).
No-Code AI Agent Platform
Full-stack platform allowing users to create custom AI assistants, connect them to structured data, and deploy them via an embeddable iframe widget with real-time WebSocket communication. Dynamic Tool Generation pipeline using LLM structured outputs.
Automate Website Testing with AI
AI-powered web testing automation tool using Playwright and LLMs to generate and execute test cases from natural language.
Authoring paper on "Dynamic Tool Generation" introducing a novel pipeline that autonomously converts raw data into executable LLM tools, validated to cut development time by ~90%.
Developed and deployed 10+ freelance projects end-to-end — from initial client consultation and system design to production deployment and ongoing maintenance, taking full ownership of the entire development lifecycle across diverse tech stacks.
Sir M. Visvesvaraya Institute of Technology, Bengaluru
Expected Aug 2027 · CGPA: 8.5/10
One more thing...
Click to find out
AI, voice agents & full-stack engineering
I Run My Company From My Watch Now. Here's How to Build It.
A beginner's guide to running your Hermes or OpenClaw agent by voice from an Apple Watch. One Apple Shortcut, a Cloudflare tunnel, a tiny Flask bridge, and async notifications.
Why Hermes Can't Beat OpenClaw
Hermes wins on memory. But multi-agent is where OpenClaw is genuinely ahead, and the gap is architectural, not cosmetic.
I Spent Weeks Debugging LangGraph. Arya Fixed It Out of the Box.
Shared state, half-written crashes, free-text agent outputs. The problems that took weeks to fix in LangGraph ship as defaults in Sarvam Arya.
