AI
for
Science
&
Engineering
Workshop
February 12, 2026
Yardi School of Artificial Intelligence, IIT Delhi
Yardi School of Artificial Intelligence, IIT Delhi
Shiv has been appointed by Hon'ble Prime Minister of India as CEO, Anusandhan National Research Foundation (ANRF). He was previously CTO, Energy Industry, Asia at Microsoft. Previously he was Executive General Manager of Growth Offerings at GE Power Conversion responsible for new Line of Business development in e-Mobility, Commercial & Industrial Solar and digital/AI innovations. Earlier he was at IBM Research - India, and the Chief Scientist of IBM Research - Australia. Before IBM, he was a tenured Full Professor at Rensselaer Polytechnic Institute in Troy, NY, USA. Shiv has degrees from Indian Institute of Technology, Madras (B.Tech, CS), Ohio State University (MS, PhD) and RPI (Executive MBA). Shiv is a Distinguished Alumnus Awardee of IIT Madras (2021, recognizing 0.3% of IITM's alumni over the years) & Ohio State University (2021), Fellow of the IEEE (2010), Fellow of Indian National Academy of Engineering (2015), ACM Distinguished Scientist (2010), Microsoft Gold Club (2024), MIT Technology Review TR100 young innovator (1999).
Anusandhan National Research Foundation (ANRF) has been recently established as a statutory body with a governing board chaired by the Hon’ ble Prime Minister. This body is aimed at driving a significant transformation in Research and Innovation across all stakeholders in India. This includes all of government, academia (public, private), industry, startups, CSR, Foundations and beyond. Collaboration across disciplines, teams vs individuals, institutions, international and quality / impact will be a major focus. The talk will also outline the vision, principles and high level approaches ANRF will use to drive collaborations and co investments across stakeholders. The talk will also include an overview of ANRF is operationalizing its grants program (₹50000 CR over 5 years, including partnerships and mission-mode programs) and the RDI (research, development and innovation) capital program (₹1 lakh CR over 5 years). We invite deeper community participation from all stakeholders, and especially industry, academia, labs, foundations/philanthrophy to come together with ANRF on the journey to Viksit Bharat.
Kamal Das is a Senior Research Scientist and Technical Assistant to the Director, IBM Research India, and CTO, IBM India & South Asia. With expertise in geospatial data, AI, and high-performance computing (HPC), Kamal has led several pioneering initiatives, including the 'Prithvi' Geospatial Foundation Model (developed in collaboration with NASA) and the High-Resolution Daily Soil Moisture (HDSM) service that has advanced precision agriculture practices at scale. He has co-authored over 80+ research papers, filed more than 15+ USPTO patents, and received multiple honors such as the Young Scientist Award, Aegis Graham Bell Award, and the IBM Outstanding Technical Achievement Award. Kamal actively contributes to industry research in geoscience, agriculture, climate, and sustainability, collaborates with global partners.
Rising atmospheric CO2 is a major driver of global warming, making timely and accurate emission monitoring essential. Identifying sources such as oil and gas facilities, urban areas, and power plants is key to effective mitigation. In this talk, I present a high-resolution, top-down framework that combines satellite remote sensing and AI to estimate CO2 emissions. A machine-learning model integrates satellite CO2, weather, and human activity data to produce daily, high-resolution concentration maps. Background CO2 is then separated from weak anthropogenic enhancements using spatial segmentation and co-emitted tracers such as NO2, and these enhancements are converted into emission rates using a physics-based mass-balance approach. Results align closely with official annual CO2 inventories. Unlike traditional bottom-up methods, this approach provides near-real-time, continuous emission estimates across multiple sectors, including oil and gas.
Dr. Manish Modani has nearly two decades of experience and has made pioneering contributions in hybrid computing, with impactful work spanning Generative AI, weather and climate sciences, quantum technologies, and high‑performance computing (HPC). He has played a significant role in several major national missions, including the National Supercomputing Mission, Mission Mausam, the National Quantum Mission, IndiaAI, and ANRF, and has actively contributed to strategic collaborations and community‑driven initiatives. Dr. Modani is currently a Principal Solution Architect at NVIDIA Graphics Pvt. Ltd. Prior to joining NVIDIA, he spent a decade at IBM, where he served as Technical Lead for The Weather Company for India and South Asia. He holds a PhD from the Centre for Atmospheric Sciences, IIT Delhi, and his research contributions have been published in multiple international peer‑reviewed journals.
Accelerated computing, combined with advances in Artificial Intelligence (AI), is transforming scientific discovery by enabling faster and more scalable analysis of complex systems. By integrating AI with heterogeneous computing architectures such as GPUs, specialized accelerators, and high‑performance computing (HPC) platforms, scientific workflows can overcome traditional constraints of computational cost and time‑to‑solution. AI for Science leverages these capabilities to process large, high‑dimensional datasets, using physics‑informed and hybrid AI–simulation approaches to improve efficiency while maintaining scientific rigor. This talk highlights key challenges in Environmental and Climate Sciences and presents generative‑AI–enabled, accelerated computing solutions, illustrated through India‑specific use cases for climate risk, sustainability, and resilience.
Nipun Batra is an Associate Professor in Computer Science at IIT Gandhinagar, where he leads the Sustainability Lab. He previously completed his postdoc from University of Virginia and his PhD from IIIT Delhi as a TCS PhD fellow. His group develops AI-powered solutions for critical sustainability challenges including smart buildings, air quality monitoring, and wearable healthcare technologies. His work has been awarded several awards, including ACM eEnergy Test of Time Award 2025, ACM SigEnergy Rising Star Award 2025, Excellence in Teaching Award IITGN 2024, Young Alumni Award IIIT Delhi 2023, Best PhD Presentation ACM SenSys 2015, Best Demo ACM BuildSys 2014, and Best Video Nominee ACM KDD 2016.
Air pollution in India is a multi-scale systems problem, shaped by emission sources, atmospheric dynamics, sparse sensing, and high barriers to interpretation. This talk presents our efforts to build an end-to-end AI decision support ecosystem that connects satellite observation, physical sensing, atmospheric modeling, and human interaction. We develop learning- and information-theoretic methods for equitable sensor placement, satellite-based emission inventory construction through automated brick-kiln detection, and physics-informed neural emulators that reduce multi-hour atmospheric simulations to real-time what-if analysis. Finally, conversational systems such as VayuChat expose these models and datasets through accessible natural-language interfaces. Together, these components demonstrate how AI can move air-quality research from isolated models to actionable, interpretable, and participatory decision-making.
Dr. Ajay Sethi is the founder of Open Science Stack (OSS), a non-profit initiative building the digital public infrastructure for AI-powered self-driving labs. Over the past two years, OSS has been developing OSS as a platform that enables researchers to run wet-lab or real-world experiments by leveraging the best-in-class AI models for experiment execution, instrument control and robotics-based automation. OSS aims to make advanced lab capabilities accessible at scale by enabling researchers to automate and run real-world experiments across domains, labs, and instruments. Prior to OSS, Ajay spent more than a decade at Accel India, one of India's leading venture capital firms, which has invested over $2 billion in Indian startups. During his time there, he worked closely with more than 150 startups across sectors and stages, including companies such as BrowserStack, Clevertap, Juspay, Swiggy, Urban Company, and Vedantu. Ajay is also the author of Startup Calculus, which proposes a rigorous, mathematical, and methodical approach to building and scaling startups. He teaches deep-tech entrepreneurship at IISc and actively mentors and advises deep-tech startups and academic founders at IIT Bombay and IISc.
Over the next decade, scientific research will undergo a fundamental shift -- from traditional, manual experimentation to AI-powered, autonomous experimentation enabled by self-driving labs. These labs will use AI to manage key parts of the experimental workflow, dramatically improving both the throughput, repeatability, and reproducibility of scientific discovery. To build self-driving labs, we need (a) foundation models across science and engineering domains that can apply scientific reasoning to generate viable hypotheses and select higher-probability experiments, (b) world models and control models that can execute experiments efficiently, safely, and consistently to maximize throughput and repeatability, and (c) action models that enable flexible and extensible robotic operations across diverse lab environments and scientific equipment while ensuring reproducible execution. Open Science Stack (OSS) is a non-profit initiative that is building the platforms needed to enable AI-powered self-driving labs. OSS is developing infrastructure that cleanly separates intent, execution, and control, while allowing AI to operate at the right level and time scale. By integrating structured experiment specifications, AI-driven reasoning, and robotic actuation, OSS enables autonomous feedback loops that continuously refine scientific workflows. OSS supports high-throughput, safe, and repeatable experiment execution at scale across domains, lab environments, and instruments, enabling reproducible outcomes across labs and researchers. By empowering researchers to automate real-world experiments, OSS aims to democratize access to advanced research infrastructure and accelerate scientific discovery.
Sai Gautam Gopalakrishnan is an Associate Professor of Materials Engineering at the Indian Institute of Science (IISc), Bengaluru. Sai obtained his dual degree (B.Tech.+M.Tech.) in Metallurgical and Materials Engineering, Indian Institute of Technology Madras, and did his PhD in Materials Science and Engineering at the Massachusetts Institute of Technology. Before joining IISc, Sai was a post-doctoral scholar in Mechanical and Aerospace Engineering at Princeton University. Sai's research interests are in the use of computational and informatics techniques to advance materials design and optimisation with applications in the energy storage and energy harvesting domains. Sai is a recipient of numerous awards including the Manohar Parrikar Yuva Scientist Award by the Government of Goa (2025), the Young Engineer Award by the Indian National Academy of Engineering (2025), the Young Metallurgist Award by the Ministry of Steel (2023), the Young Battery Researcher Award (2023), and the Prof. Priti Shankar Teaching Award (2023).
The performance of most engineering applications, ranging from infrastructure, semiconductor devices, and energy is critically limited by the materials that are deployed and their associated properties. Hence, significant gains in performance of any application can arise from the usage of 'better' materials. However, designing better materials is historically a 'slow' process, limited by the throughput of experiments, calculations, and the general availability of data. Thus, machine learning (ML) and artificial intelligence (AI) tools, particularly those that work well in 'small data' domains can play a crucial role in accelerating the innovation at the materials level, thereby improving the devices that we use. In my talk, I will provide an overview of the typical AI/ML tools that are used (so far) in materials research and design, including examples from our own work. I will conclude with a few open-ended points that are still up for discussion in the use of AI/ML for materials design.
Dr. Chandan Kumar Choudhury is a computational chemist with expertise in quantum and molecular simulations, as well as AI-based drug and materials discovery. He earned his PhD in 2016 in Computational Chemistry from CSIR-National Chemical Laboratory (NCL), Pune, and subsequently completed postdoctoral research at Clemson University, USA. Since 2020, Dr. Choudhury has been an integral member of Prescience Insilico Pvt. Ltd. (Bengaluru), a deep-tech company focused on expediting drug and materials innovation through artificial intelligence, machine learning, and physics-based multiscale simulations. In his current role as Manager of Material Science R&D, he leads the application of computational methodologies for the design of advanced materials, encompassing alloys, polymers, formulations, and sustainable solutions. Dr. Choudhury has authored 20 publications in peer-reviewed journals. With robust training in computational chemistry and extensive experience on both academic and industrial fronts, he is dedicated to advancing computational tools that drive meaningful progress in materials science and pharmaceuticals.
Running sophisticated molecular simulations has traditionally required years of specialized training. A graduate student might spend their first year just learning how to set up a single type of calculation correctly, with subtle errors in input files leading to days of wasted computing time. This creates a bottleneck where even brilliant scientists with important questions must either become simulation experts themselves or depend entirely on the few specialists who are. Aura aims to change this by putting AI between researchers and the complexity. It's a conversational system where you describe your scientific questions such as how a new solar cell material conducts charges, how a potential drug binds to its target, or how a soap molecule organizes in water and intelligent agents translate that into the dozens of precise technical steps required. The system handles the tedious work of file format conversions, parameter selection, and error detection, while keeping researchers informed about possible underlying reasons. This doesn't replace expertise, but it dramatically lowers the barrier to getting started and reduces the cost of mistakes.
Jay Shah is a Senior Quantum Machine Learning Engineer at BQP - a startup focused on using Quantum Computing for Aerospace and Defense with headquarters in New York. He is the Founder of QRL that launched qrl-qai python SDK as an attempt to create quantum analogue of OpenAI’s gym framework. He is also a podcast host, Quantum Tech YouTuber and Quantum Educator. His work focuses on the applications of Quantum Machine Learning in Engineering Simulations and Quantum Reinforcement Learning. He is an active open-source contributor and has created projects like “Quantum Glasses” software - a proud member of the IBM Qiskit Ecosystem. On his podcast show “Quantum Podcast with Jay Shah”, he hosts podcasts with leading quantum computing professionals to discuss the evolving quantum landscape. He is a quantum educator who shares quantum computing-based educational content on his YouTube channel “QRL”. He has been a speaker/resource person at 14+ Universities including Arizona State University, IIT Delhi, IIT Jodhpur, IIT Guwahati, IIIT Bangalore, and NIT Agartala.
The solution of partial differential equations (PDEs) lies at the heart of scientific computing, governing phenomena across physics, engineering, and finance. In recent years, machine learning–based approaches have emerged as powerful alternatives to traditional numerical solvers, offering mesh-free formulations, improved scalability, and data-driven generalization.This talk presents a unified overview of classical, quantum, and hybrid AI models for solving PDEs. We begin with Physics-Informed Neural Networks (PINNs), discussing their formulation, training dynamics, and practical challenges such as optimization stiffness and spectral bias. Building upon this foundation, we explore operator-learning frameworksincluding Deep Operator Networks (DeepONets) and Physics-Informed Neural Operators (PINOs), highlighting their ability to learn solution operators and generalize across families of PDEs and boundary conditions.The talk then transitions to the quantum domain, introducing Quantum-Assisted Physics-Informed Neural Networks (QAPINNs) and Quantum DeepONets as a hybrid paradigm that integrates quantum computing within physics-informed learning pipelines. We discuss how quantum representations and hybrid quantum–classical training can potentially enhance their classical counterparts.
Prof. Sashikumaar Ganesan is a pioneer in Scientific Machine Learning (SciML) and Physics-AI Integration. He leads the AI for Research and Engineering eXcellence (AiREX) lab at IISc, where his team builds the mathematical backbones for next-generation AI. Bridging academia and national strategy, he is the Founder of ZenteiQ AiTech Innovations. Under his leadership, ZenteiQ was selected by the Government of India (IndiaAI Mission) to build India's first Sovereign Scientific Foundation Models. ZenteiQ is building a portfolio of 8-80 billion parameter Scientific Foundation Models designed to accelerate R&D in aerospace, defence, and energy.
SciREX is an open-source scientific AI framework combining Physics-Informed Neural Networks and Fourier Neural Operators for solving PDEs across engineering domains. Built with hardware-aware optimization and reproducible workflows, SciREX delivers physics-validated, mathematically interpretable predictions for fluid dynamics, thermal analysis, and electromagnetics. We demonstrate applications spanning convection-dominated flows and design optimization, showing how open-source scientific computing tools accelerate industrial R&D while preserving rigor and reproducibility.
Dr. Shirish Karande is currently working as Principal Scientist and Head of Media and Advertising Research Area at TCS Research. He got his Ph.D. in Electrical and Computer Engineering from Michigan State University in 2007. He has been with TCS Research for last 16+ years, where his work has spanned various Applied AI challenges leading to 30+ granted patents. His contributions to Scientific Machine Learning span across building neural surrogates for Industrial systems, use of hypernetworks or neural operators as fast solvers, symbolic and neuro-symbolic solutions for PDEs, use of LLMs for accelerating research and data science workflows. His current broad interest in AI are focused on AI Alignment, AI Creativity and Self Improvement in AI.
PDE-based modeling underpins a wide range of scientific and engineering work at TCS Research, from industrial systems and materials to biological and climate modeling. Using neural twins for industrial systems as a concrete reference point, this talk reflects on how compute, creativity, and trust bottlenecks shape our ongoing research in AI for PDE-driven computation, and motivates broader future directions in scientific machine learning.