I am a Computer Science & Engineering undergraduate at the National Institute of Technology, Hamirpur (Expected Graduation: 2028). My primary technical focus lies in Embodied AI, Vision-Language-Action (VLA) Models, 3D Computer Vision, and Robotics Simulation. I specialize in bridging reality-to-simulation gaps, optimizing deep learning inference pipelines, and engineering physics-constrained graph neural network architectures.

• Languages: Python, C++
• Deep Learning & Vision: PyTorch, JAX, TensorFlow, PyTorch Geometric, Computer Vision, Vision-Language Models (VLMs), Vision-Language-Action Models (VLAs)
• Optimization & Deployment: CUDA, TensorRT, ONNX, Inference Optimization
• Robotics & Simulation: ROS2, Imitation Learning, Motion Retargeting
• MLOps & Infrastructure: Docker, Git, GitHub, MLflow, Weights & Biases (W&B)
• Data & Scientific Computing: NumPy, Pandas, OpenCV, Scikit-learn, Zarr

Reality-to-Simulation Pipeline for Embodied AI | Dec 2025 – Mar 2026

• Developed high-fidelity robotic training environments directly from video data to bridge the reality-to-simulation gap in Embodied AI.
• Engineered a monocular 3D reconstruction pipeline integrating depth estimation and geometric reasoning.
• Implemented graph-based camera calibration to achieve precise positional alignment across coordinate frames.
• Built human-to-robot motion retargeting pipelines to map human demonstration data onto robotic embodiments while preserving kinematic constraints.

Mini JAX-Based Vision-Language-Action Model | Aug 2025 – Jun 2026

• Developed and trained a Vision-Language-Action (VLA) model from scratch in JAX, scaling to 1M parameters across 600+ robotic demonstration datasets.
• Implemented patch-based action tokenization, allowing the model to operate seamlessly across robot action spaces of arbitrary dimensionality.
• Enabled long-horizon robotic control with action prediction horizons exceeding 50 timesteps for manipulation and navigation tasks.
• Optimized training and inference pipelines to achieve a real-time control latency of 50 ms.

Surrogate Model for Electrical Grids | Aug 2025 – Jun 2026

• Built a Graph Attention Transformer surrogate model to learn grid-wide power allocation directly from network topology and operational conditions.
• Formulated a physics-constrained objective function featuring 10+ differentiable loss terms to enforce physical feasibility, stability, and grid operational limits during training.
• Trained on solver-generated optimal trajectories to deliver reliable near-optimal power allocations where non-convex optimization methods struggle.
• Reduced inference time by 5× compared to standard solvers while maintaining solution fidelity on IEEE benchmark power systems.

National Institute of Technology, Hamirpur
B.Tech in Computer Science & Engineering | 2024 – 2028

• CGPA: 9.12 / 10.0
• Relevant Coursework: Data Structures & Algorithms, Operating Systems, Computer Architecture, Probability & Statistics, Discrete Mathematics, Python Programming.

• Hackathon Organizer: Organized 2 national-level hackathons with 1,000+ total registrations, managing technical evaluation and industry sponsorships.
• Executive, CSEC (Computer Science Engineers Community, NIT Hamirpur): Spearheaded technical events, coding contests, and peer mentoring programs for student developers.
• Executive, GDG (Google Developer Group, NIT Hamirpur): Developed open-source AI projects and conducted hands-on technical workshops covering deep learning fundamentals and modern frameworks.