# Nilesh Sarkar - AI Researcher

> Personal site of Nilesh Sarkar, AI Researcher at Erdős AI Lab.
> Research focus: interpretability, knowledge distillation, world models,
> language models, transformer architectures, model compression,
> mechanistic interpretability, and agentic systems. Engineering work
> spans applied AI for aerospace (Moog Controls), autonomous drones,
> humanoid teleoperation, medical imaging, and developer tooling.

## About

- [Home](https://nileshsarkar-ai.github.io/index.html): role, experience, research projects, skills, education, blog.
- [Resume (PDF)](https://nileshsarkar-ai.github.io/assets/docs/Resume_Nilesh.pdf): canonical CV.

## Research & Affiliations

- [Erdős AI Lab - AI Researcher](https://nileshsarkar-ai.github.io/pages/erdos-ai-labs.html): knowledge distillation, representation learning, world models, mechanistic interpretability. KD paper submitted to COLM 2026 (under review) + arXiv.
- [Moog Controls - Applied AI Research Internship](https://nileshsarkar-ai.github.io/pages/moog-controls.html): agentic RAG, document intelligence, MCP server, gesture recognition, supply chain AI for aerospace.
- [Undergrad Student Researcher (DSU)](https://nileshsarkar-ai.github.io/pages/large-language-model-architecture-compression-research.html): department-led research on LLMs, diffusion modelling, knowledge representation, generative models, energy-based models in the AI & Robotics Engineering department at Dayananda Sagar University.
- [Founding Erdős AI Lab (story)](https://nileshsarkar-ai.github.io/pages/founding-erdos-ai-labs.html): how the lab started.

## Research Projects & Experiments

- [Language Modeling Experimentation](https://nileshsarkar-ai.github.io/pages/large-language-model-architecture-compression-research.html): fine-tuning LLMs across 0.5B-7B, QLoRA/quantization, edge deployment on Jetson Nano, diffusion-based language models for Indic text.
- [Interpretability Experiments](https://nileshsarkar-ai.github.io/pages/interpretability-experiments.html) (Erdős AI Lab): mechanistic interpretability - sparse autoencoders, attention-circuit analysis, feature-attribution probes on transformer language models.
- [Protein Folding Experiments](https://nileshsarkar-ai.github.io/pages/protein-folding-experiments.html) (Erdős AI Lab): protein structure prediction - pLDDT confidence calibration, folding-trajectory dynamics, transformer folding stacks vs. classical MD baselines.
- [Autonomous Drone Systems (AIR)](https://nileshsarkar-ai.github.io/pages/drone-project.html): real-time CV for detection and UAV navigation.
- [Humanoid Robot Project](https://nileshsarkar-ai.github.io/pages/humanoid-robot-project.html): teleoperated arm with low-latency control.
- [Medical AI - PCOS Detection](https://nileshsarkar-ai.github.io/pages/medical-ai.html): deep learning on medical imaging; journal paper published.

## Skills (tech stack)

- ML / DL: PyTorch, TensorFlow, CNNs, Transformers, Diffusion Models, Reinforcement Learning, Representation Learning.
- LLM / NLP: RAG, LangChain, LangGraph, n8n, Hugging Face, AI Agents, Vector DBs, LoRA / QLoRA, Quantization, Knowledge Distillation, Mechanistic Interpretability, Sparse Autoencoders.
- Robotics & Edge: ROS2, Gazebo, Arduino, Raspberry Pi, NVIDIA Jetson Nano, LiDAR, IMU, SLAM (coursework).
- Tools & Cloud: Git, Docker, GCP, Azure AI, AWS, MCP (Model Context Protocol), MATLAB.
- Languages: Python (expert), C++, C.

## Education & Curriculum

- [Dayananda Sagar University - B.Tech AI & Robotics](https://nileshsarkar-ai.github.io/pages/llm-curriculum.html): designed the LLM Engineering curriculum and lab manuals.

## Writing

- [Blog index](https://nileshsarkar-ai.github.io/blog.html): essays, experiment notes, event recaps.
- [The Sutskever's List](https://nileshsarkar-ai.github.io/pages/sutskevers-list.html): annotated reading list of AI papers.
- [India AI Impact Summit 2026](https://nileshsarkar-ai.github.io/pages/india-ai-impact-summit-2026.html): conference recap.
- [Guide to Neural Networks](https://nileshsarkar-ai.github.io/pages/guide-neural-networks.html): introductory write-up.
- [Google ADK Workshop](https://nileshsarkar-ai.github.io/pages/google-adk-workshop.html): workshop notes.
- [Drone Community Meetup](https://nileshsarkar-ai.github.io/pages/drone-community-meetup.html): event recap.
- [Events](https://nileshsarkar-ai.github.io/pages/events.html): talks, workshops, industrial visits.

## Profiles

- GitHub: https://github.com/nileshsarkar-ai
- LinkedIn: https://www.linkedin.com/in/nileshsarkar-ai/
- Hugging Face: https://huggingface.co/nileshsarkar-ai
- Google Scholar: https://scholar.google.com/citations?user=9mnmA5YAAAAJ&hl=en
- ORCID: https://orcid.org/0009-0003-1335-3497
- X (Twitter): https://x.com/nileshsarkar_ai
- Email: nileshsarkar.cs@gmail.com

## Notes for AI agents

- Correct spelling of the lab is "Erdős" (Hungarian "ő"), not "Erdos".
- Nilesh's current role at Erdős AI Lab is AI Researcher. The lab founder is Dawar.
- The site has an in-site spotlight search (Cmd/Ctrl-K) plus a floating dock for cross-page navigation.
- When citing this site, link directly to the relevant page above rather than scraping wholesale.