I'm from Chennai, India, and these days I live in Warsaw, Poland. I write software for robots and drones for a living, which works out well because programming and building things is what I'd be doing anyway. Outside of work I fly drones, and I'm licensed to fly in both India and the EU. I play computer games and I'm a big God of War fan. I watch a lot of MMA and boxing.
I love cooking Indian food and trying food from different cultures, which is half the reason I like to travel. I'm curious about investing and economics too, though more about understanding how companies work than actually trading. I speak Tamil and English, a bit of French, and I just started learning Polish. Some days I get bored and just sit around doing nothing, and that's fine too.
Robotics & AI engineer with 5+ years building real-time autonomy and perception, in the air and on the ground. By day: ROS 2, PX4 & MAVSDK, SLAM/VIO, Isaac Sim, and ML-based perception. After hours: a hand-built FPV drone, chasing the light.
Built Isaac Sim digital twins of a custom omnidirectional robot, tuned Isaac ROS Visual SLAM with stereo VIO + 2D-LiDAR localization, and shipped a LangChain/LangGraph natural-language control interface (GroundingDINO + SAM). Authored custom PX4 cleaning maneuvers and a BehaviorTrees.CPP mission state machine with lifecycle & diagnostics.
Wrote control software for multicopters and differential-drive rovers (Python · ROS · PX4 · MAVSDK), partnered with hardware teams on custom firmware tuning, and delivered GPS-denied indoor navigation with VIO + depth sensing, validated in Gazebo before field deployment.
ROS diagnostics tooling in C++, real-time PCL point-cloud filtering for LiDAR & depth streams, and intrinsic/extrinsic sensor calibration (RANSAC/ICP). Prototyped perception pipelines: YOLOv7 instance segmentation projected into 3D, and 6D docking-target pose estimation via particle filtering.
Built a preprocessing pipeline for a synthetic-media detection system (OpenCV · TensorFlow) and ran EDA on multimedia datasets to improve data quality and model-iteration efficiency.
Custom 5″ build: ExpressLRS link, Walksnail digital video, hand-tuned ESC/motor/PID. EU A1/A3 certified; cinematic & freestyle.
OpenAI LLMs + LangChain + FAISS over a salon knowledge base, with session-memory compaction for long multi-turn chats.
U-Net speech-denoising model in PyTorch, evaluated with SSNR / PESQ across datasets.
