21 · Engineer · Researcher
I build AI agent infrastructure — agent runtimes, memory systems, and multi-agent orchestration that run locally, without a proprietary cloud SDK. I contribute to open-source AI infrastructure (mem0, OpenClaw) and have shipped production Web3 and IoT systems.
◆ current work
- NeoSwarm — local multi-agent orchestrator: a team-lead agent delegating to parallel workers, native Python agent loop, zero cloud SDK.
- OpenLog — AI-native release manager that turns raw commits into polished changelogs (Groq + BlockNote).
- AgentReady Commerce — merchant AI sales agent that binds approvals to exact commercial terms before completing a Razorpay transaction.
- Evo2 Variant Analysis — decentralized genomic variant analysis on Google's Evo 2 + BNB Chain. BNB Hack Kerala winner.
- Quack — AI rubber duck: a Rust terminal companion that explains errors without breaking flow.
- Enduride — IoT telematics cutting crash-alert delay to under 5s (Arduino + Solidity).
◆ open source
- mem0 — merged fix mapping entity params to filters in the memories API (LLM memory layer).
- OpenClaw — merged fixes to Ollama model discovery and the config pipeline (agent orchestration).
- Open Design — merged fix for blocked PDF export across web and desktop.
◆ my usual stack
- python · typescript · rust · go · c++ · swift · react native · solidity
◆ workflow I build with agents as much as I build them. My loop: write a tight spec → delegate the draft to coding agents → review every diff → back it with tests. Agents are fast junior engineers, not authors.
- agent stack: monocode · pi · opencode
◆ achievements & leadership
- Winner — TrackShift 2026 (2nd place, national) — Mphasis × TGR Haas F1 Team innovation challenge (180+ students). Built an F1 telemetry replay with a hierarchical AI race engineer.
- IEEE-published — ThinkDiff-SLM, a masked-diffusion language model for on-device edge deployment.
- Hackathon Veteran — Winner of Enduraverse (₹100K), BNB Hack Kerala, HackGenX, and SVPCET Hack-A-Thon. SIH 2025 finalist; finalist in 15+ national events.
- IEEE CS @ YCCE — Vice Chair, community of 300+ members.
- Codeware — Chairperson, 500+ members.
◆ how i work I build the layer under the chat. Most "AI apps" are a prompt wrapped around an API call — I care about the runtime, the memory, and the orchestration that make an agent survive real use. I live in the terminal on Arch (Omarchy + Hyprland) — keyboard-first, optimized for system-level control. I'd rather ship a working system than win an argument about architecture.
◆ currently exploring
- agent infrastructure — memory architectures, long-running agents, tool-calling reliability
- systems programming — Rust, C++, high-performance CLI tooling
- edge AI — running small language models locally, where the cloud can't reach
◆ beyond work
- Formula 1 & motorsport — race strategy, telemetry, and vehicle dynamics. Competed in the TGR Haas F1 Team–backed TrackShift 2026 challenge.
- Aerospace & private space — long-term interest in launch systems, satellites, and space infrastructure.