Nilesh Sarkar

AIR Research Group: Drone Project Research Group
I am a researcher within the AIR Research Group, focusing on the development of autonomous drone systems. My work involves implementing computer vision algorithms for real-time object detection and navigation, contributing to the group's broader goals of deploying intelligent UAVs in complex environments.
Robotics Computer Vision UAV

Sim-to-Real Humanoid Control Ongoing
Developing robust locomotion policies for a miniature humanoid robot using Deep Reinforcement Learning (PPO) in NVIDIA Isaac Sim.
Focusing on domain randomization to bridge the sim-to-real gap, allowing the policy to handle uneven terrain and external perturbations on physical hardware.
Reinforcement Learning Isaac Sim Omniverse Sim-to-Real

Moog Controls : Applied AI Research Intern
I am working as an Intern at Moog Controls, Moog India Technology Center (MITC), Bengaluru, where I have been involved in developing AI-driven, safety-critical systems using large language models (LLMs). My work focuses on retrieval-augmented generation (RAG) pipelines built with LangChain and LangGraph, with an emphasis on improving reliability, traceability, and real-world deployability of LLM-based systems in high-stakes engineering environments.

During my initial two-month internship, I worked on refining internal RAG systems, improving retrieval accuracy from approximately 70% to over 80%, and extending support to complex and multi-modal technical documents. I also contributed to early-stage work on LLM-orchestrated CAD generation and automated engineering analysis workflows.

Following this, I was granted an additional three-month internship extension to continue working on AI agents, AI applications in supply-chain workflows, and generative AI development based on transformer and attention mechanisms, supporting broader adoption of AI across engineering and operational domains.
RAG Generative CAD Safety-Critical

Large Language Model Architecture & Compression Research (Ongoing Research Project) Research
I am currently working on a research project focused on understanding and designing efficient large language model (LLM) architectures for constrained and real-world deployment settings. The project adopts an architecture-first research approach, studying how transformer design choices influence representation quality, robustness, and efficiency under limited computational budgets. My work uses training of compact transformer models and compression as an analytical tool to study information bottlenecks and architectural redundancy.
LLM Efficiency PyTorch CUDA Model Compression

Serving as a Student Executive Committee Member for IEEE RAS CIS SBC DSU, I lead the organization of technical workshops, expert talks, and hackathons centered on AI and robotics, fostering a collaborative learning environment for peers.

As a Core Committee Member for Roboverse, I actively coordinate and conduct hands-on workshops and hackathons, helping to bridge the gap between theoretical concepts and practical application in AI and robotics.

Dayananda Sagar University, Bangalore
Undergraduate (B.Tech) in Artificial Intelligence and Robotics Engineering

I have actively contributed to the academic framework of the department. I played a key role in designing the curriculum, lab manuals, and coursework for the LLM Engineering syllabus. This initiative involved structuring a comprehensive learning path that bridges theoretical fundamentals with practical implementation, enabling peers to gain hands-on experience in building, fine-tuning, and deploying large language models.
Computer Vision LLM/ML/AI Deep Learning ROS Robotics CNN Transformers