Akshay Govind Srinivasan

I am Akshay, a Final-year Bachelors student majoring in Mechanical Engineering and Minoring in AI at the Indian Institute of Technology Madras (IITM) and an incoming PhD Student at Massachusetts Institute of Technology in the CCSE-MechE joint programme.

Research Interests: Vast majority of computational power in the world is used to understand, design and optimize complex scientific and engineering systems like cars, airplanes etc. My research goal is to use AI to accelerate and democratize this computation to (1) achieve similar accuracy with much less compute, (2) solve problems that were computationally impossible, and (3) democratize the usage of these tools by reducing the knowledge barrier required for it’s effective usage.

At IIT Madras, I was advised by Prof. Balaji Srinivasan (WSAI, IITM) on building Physics-Informed Extreme Learning Machines (PI-ELMs), a fast, intrepretable and sustainable alternative to Physics-Informed Neural Networks. Previous to this, I have worked on introducing physics constraints into deep learning models for inverse airfoil design under guidance of Prof. Nagabhushana Rao Vadlamani and Prof. Bharath Govindarajan. I was also fortunate to have worked under the mentorship of Prof. Balaraman Ravindran and Dr. Gokul S Krishnan at Center for Responsible AI (CERAI) in building Robust Bias Evaluation Frameworks.

news

Apr 10, 2026	My first patent has been officially granted!
Jan 10, 2026	Honored to receive two awards at AAAI2026 for our research on LLM bias evaluation.
Dec 16, 2025	Our paper "Towards Sustainable Scientific Machine Learning Fast and Interpretable PDE Solvers via RBF-PIELM" has been accepted at ACM CODS 2025 as oral and archival work !
Jul 18, 2025	Our paper “IndiCASA: A Dataset and Bias Evaluation Framework in LLMs Using Contrastive Embedding Similarity in the Indian Context” has been accepted at AAAI/ACM Conference on AI, Ethics, and Society (AIES 2025) !
May 19, 2025	I have joined Adobe Media and Data Science Research Lab as an intern working on building Multi-Agent Collaborative RL framework for Video Tasks

selected publications

Beyond the Moment: Conditioning Frozen VLAs on Memory for Long-Horizon Manipulation Tasks

Aditya Ramesh, Jatin Chauhan, Akshay Govind Srinivasan, and 2 more authors

In ICML 2026 Workshop on Multimodal AI Agents, 2026

Bib

@inproceedings{ramesh2026beyond,
  title = {Beyond the Moment: Conditioning Frozen {VLA}s on Memory for Long-Horizon Manipulation Tasks},
  author = {Ramesh, Aditya and Chauhan, Jatin and Srinivasan, Akshay Govind and Bhardwaj, Shivam and Kaul, Manohar},
  booktitle = {ICML 2026 Workshop on Multimodal AI Agents},
  year = {2026},
  url = {https://openreview.net/forum?id=F9w2kLRL8a},
}

Enhancing Financial RAG with Agentic AI and Multi-HyDE: A Novel Approach to Knowledge Retrieval and Hallucination Reduction

R. George, A. G. Srinivasan, J. K. Joe, and 4 more authors

In Proceedings of The 10th Workshop on Financial Technology and Natural Language Processing, 2025

Bib

@inproceedings{george2025enhancing,
  title = {Enhancing Financial RAG with Agentic AI and Multi-HyDE: A Novel Approach to Knowledge Retrieval and Hallucination Reduction},
  author = {George, R. and Srinivasan, A. G. and Joe, J. K. and MR, H. and Kant, H. and Vimalkanth, R. and Suresh, S.},
  booktitle = {Proceedings of The 10th Workshop on Financial Technology and Natural Language Processing},
  year = {2025},
}

Deep vs. Shallow: Benchmarking Physics-Informed Neural Architectures on the Biharmonic Equation

A. G. Srinivasan, V. Dwivedi, and B. Srinivasan

In Machine Learning and Physical Sciences Workshop at the 39th Conference on Neural Information Processing Systems (NeurIPS), 2025

Bib

@inproceedings{srinivasan2025deep,
  title = {Deep vs. Shallow: Benchmarking Physics-Informed Neural Architectures on the Biharmonic Equation},
  author = {Srinivasan, A. G. and Dwivedi, V. and Srinivasan, B.},
  booktitle = {Machine Learning and Physical Sciences Workshop at the 39th Conference on Neural Information Processing Systems (NeurIPS)},
  year = {2025},
}