Amey Varhade

I am a research fellow at Microsoft Research, where I am advised by Dr. Ravishankar Krishnaswamy, Dr. Navin Goyal and Dr. Kirankumar Shiragur working broadly on problems facing retrieval; improving, evaluating, and benchmarking classical and agentic retrieval systems. More specifically, I work on algorithmic enhancements for Approximate Nearest Neighbour (ANN) based search and indexing, including improvements to Microsoft's DiskANN algorithm. I have worked on generating synthetic data for workplace communications, documents, and artifacts, which is used to evaluate and fine-tune the M365 Copilot models.

I have completed my B.Tech in Computer Science and Engineering with a minor in Product Design at Indian Institute of Technology (IIT) Guwahati. Before joining MSR, I gained industry experience at IBM India Systems Development Labs, working on IBM's Virtual Private Cloud (VPC) offerings. I also had a brief collaboration with IBM Research on Retrieval-Augmented Generation (RAG) techniques for code-generative models used in Ansible.

My main research interests span Algorithms, Databases, Learning Theory, Scientific Machine Learning/AI4Science, Interpretability and LLMs. I am interested in combining theoretical insights with practical applications to both scientific understanding and real-world impact.

I am also interested in domains related to computational efficiency in AI systems and scalable machine learning infrastructure.

I am actively looking for PhD opportunities for Fall 2026 in Computer Science. I'm open for and would welcome discussions with prospective advisors. Please feel free to reach out if my background and interests match!

Email / CV / Google Scholar / Github / LinkedIn / Work

News

[07/2025]	Paper on RI-MAC: Optimising MAC Operation using Custom RISC-V Instruction Set for Neural Network Inference accepted at AIMLSys, 2025.
[09/2024]	Excited to start as a research fellow at Microsoft Research
[12/2023]	Paper on Can Physics informed Neural Operators self improve? accepted as spotlight at DLDE-III Workshop, NeurIPS 2023.
[09/2022]	Paper on CluSpa: Computation reduction in CNN inference by exploiting Clustering and Sparsity accepted at AIMLSys, 2022.
[09/2022]	Started as Software Engineer at IBM India Systems Development Labs, working on VPC on IBM Cloud.
[06/2022]	Graduated from IIT Guwahati with B.Tech in Computer Science and Engineering.
[08/2021]	Successfully completed Google Summer of Code with ML4SCI, working on turbulent fluid dynamics analysis under Prof. Brad Marston.

Publications

	Can Physics Informed Neural Operators Self Improve? Ritam Majumdar, Amey Varhade, Shirish Karande, Lovekesh Vig DLDE-III Workshop, Neural Information Processing Systems (NeurIPS), 2023 (Spotlight Talk) Investigated self-improvement capabilities of physics-informed neural operators for scientific modeling.
	CluSpa: Computation Reduction in CNN Inference by Exploiting Clustering and Sparsity Chetan Ingle, Imljungla Longchar, Amey A Varhade, Saurabh Baranwal, Hemangee K Kapoor International Conference on AI-ML Systems (AIMLSys), 2022 Proposed a computation reduction technique for CNNs using clustering and sparsity patterns.

Collaboration & Contact

I'm always interested in collaborating on exciting research projects and discussing new ideas. If you want to discuss something research-related, or have research positions/roles that might be a good fit, please feel free to reach out via email. I'd love to hear from you!