I build intelligent systems that have to agree with the world outside the notebook.
AI Engineering undergraduate at Amrita Vishwa Vidyapeetham, working across reinforcement learning, computer vision, motion capture, and EEG biosignals. My default mode is to follow an idea all the way from paper, to prototype, to real and imperfect input data.
|
MARKERLESS MOTION INTELLIGENCE
A full offline-first motion-capture pipeline: MediaPipe pose detection, stereo DLT triangulation, filtering, joint kinematics, and production-ready exports. FastAPI React Three.js OpenCV
|
AGENTS BUILT FROM FIRST PRINCIPLES
A Tic-Tac-Toe agent trained through self-play with Double/Dueling DQN, NoisyNet exploration, and prioritized experience replay—paired with a minimax baseline. PyTorch RL self-play
|
|
BIOSIGNALS / LINEAR ALGEBRA
SVD-based EOG artifact removal for real 14-channel Emotiv EPOC X recordings: separating noise structure from neural signal without pretending the data is clean. MATLAB SVD EEG
|
OFFLINE MOBILE COMPUTER VISION
A mobile-first OMR evaluation workflow that runs entirely on device: capture, OpenCV bubble detection, answer-key matching, and exportable reports. Flutter OpenCV on-device
|
I contribute the small, sharp improvements that make other people’s software better: fixes, accessibility work, documentation accuracy, and developer experience across AI frameworks, inference tooling, and systems libraries.
| MERGED mistral-common |
Documentation correction for the experimental usage guide. | PR #306 ↗ |
| MERGED MuJoCo |
Documentation accuracy fixes for a physics-engine reference. | PR #3550 ↗ |
| ACTIVE AstryX |
IME-safe handling for TextInput.onEnter conversion commits. |
PR #6083 ↗ |
| ACTIVE NVIDIA NeMo |
Developer-experience fixes across the Run and DataDesigner toolchain. | View work ↗ |
Selected recent work · open the full contribution ledger →
