Robotics and applied-ML engineer. M.S. Mechanical Engineering (Robotics), UC San Diego · B.Tech. Mechanical Engineering, IIT Bombay.
I work on two things: learning-based autonomy in simulation, and applied-AI systems that put models in front of real users. Graduate research at UC San Diego's Existential Robotics Laboratory under Prof. Nikolay Atanasov, on reinforcement learning for autonomous navigation.
Based in India. Open to robotics, perception and applied-AI roles, remote preferred.
- Growero HRMS — HR platform in daily use by a company's employees. Next.js 16 / React 19 / TypeScript / Prisma / PostgreSQL, deployed as a PWA. Includes a full Indian payroll engine (PF, PT, ESI, TDS, attendance-driven proration, locked payroll runs), geofenced GPS attendance, leave and expense approval chains, and generated payslip and offer-letter PDFs. Built and maintained solo. Private repository — happy to walk through it.
- FarmInsta Chatbot — deployed conversational analytics for an agri field-operations team, as AI Tech Lead. Natural-language questions about farmer visits, product demonstrations and team activity are translated to SQL against a read-only copy of the operational database, with conversational context and automatic query repair. Client project — private.
- Reinforcement Learning for Autonomous Navigation — the UCSD research codebase. Custom MuJoCo / dm_control environments for an Ackermann-steered car, PPO and SAC, and a comparison of CNN, point-cloud GNN and Transformer feature extractors over depth and vehicle state.
- LocoTransformer v3 — a reproducible successor framework: Ackermann, differential-drive and skid-steer bodies across four world types, a MetaDrive adapter, four visual encoder families, CI, 40 test modules, Docker, and a versioned evaluation protocol with artifact hash contracts.
- Visual-Inertial SLAM — IMU and stereo fusion via EKF for pose estimation and landmark mapping.
- LiDAR SLAM — scan matching, occupancy grids, texture mapping and pose-graph optimisation.
- Motion Planning — search-based and sampling-based planners (A*, RRT, bidirectional RRT) in 3-D environments.
- Mobile Manipulator Control — full pick-and-place control stack for a KUKA YouBot: kinematics, trajectory generation and feedback control, simulated in CoppeliaSim.
- Comma.ai Calibration Challenge — estimating camera pitch and yaw from monocular driving video using optical flow, vanishing-point geometry and a Conv-LSTM.
- Model Predictive Control — MPC for navigation over uneven agricultural terrain with elevation and friction effects.
- windows-ipad-control — use a Windows laptop's keyboard, trackpad and gestures to control an iPad over Bluetooth LE. Universal-Control style, no extra hardware. MIT.
Python · PyTorch · NumPy · MuJoCo · dm_control · Stable-Baselines3 · OpenCV · SQL · TypeScript · Next.js · React · Prisma · PostgreSQL · Docker · MATLAB