Sashikumaar Ganesan
On academic leave from IISc from 2026, working full time on ZenteiQ
I am building frontier AI for the physical world: scientific foundation models that compute physics rather than describe it. The foundations come from my research at IISc in numerical analysis and scientific machine learning, and ZenteiQ takes them to industry.
ZenteiQ
The company I founded, and my full-time focus from 2026.
ZenteiQ AiTech Innovations takes physics-native AI into industry. It develops Dhīr, a family of scientific foundation models pre-trained from scratch rather than adapted from existing model weights, and products that bring them into engineering workflows.
- Recognition
Selected and funded under the IndiaAI Mission (MeitY) to build India's sovereign scientific foundation models. Video (opens in a new tab)- Model sizes
- 2B, 5B, 9B and 35B parameters, followed by a few trillion-parameter frontier models
- Progress
- Two Dhīr base models trained, including a mixture-of-experts model. Public release of the first two is upcoming.
- Ecosystem
- ARTPARK collaboration; member of The AI Alliance
ZenteiQ is independent of IISc. Enquiries about its models, pilots and partnerships go to ZenteiQ directly.
Visit zenteiq.ai(opens in a new tab)Approach
Why AI for physical systems has to compute, not narrate.
Language models trained on engineering text can discuss the Navier–Stokes equations fluently, but they do not solve them. Engineering decisions need numbers that hold up against a solver, a measurement or a conservation law.
My research starts from the mathematics. Finite element theory, stabilisation and error analysis say what a correct answer has to satisfy; machine learning makes reaching it fast and adaptive. At ZenteiQ, scientific foundation models carry that idea from a single equation to whole families of physical systems.
Compute the field
Models return physical quantities on a domain, the thing an engineer measures, rather than a description of it.
Check against the reference
Results are evaluated against reference numerical solutions and the physical constraints they must respect.
Build on numerical analysis
Stabilised finite elements, hp-variational PINNs and operator splitting give the learning a structure that can be analysed.
Research
AiREX lab, Department of Computational and Data Sciences, IISc.
The AI for Research and Engineering eXcellence (AiREX) lab works on scientific machine learning: combining the precision of scientific computing with the adaptability of machine learning, in theory and in applications.
Scientific machine learning themes
hp-variational physics-informed neural networks
hp-VPINNs for convection-dominated and incompressible flow problems, with tensor-based implementations for complex geometries.
The variational formulation and hp-refinement bring finite element ideas into PINN training; tensor-based assembly makes it fast on GPUs.
Neural-augmented numerical methods
Neural networks that predict stabilisation parameters inside finite element schemes, keeping the numerical method as the backbone.
Data-driven components are placed where classical schemes rely on hand-tuned parameters, so the result stays a finite element solution with its structure intact.
GPU-accelerated scientific computing
High-performance parallel computing and GPU acceleration for both learned and classical solvers.
Hardware-aware algorithms, hybrid CPU-GPU multigrid and parallel finite element software make scientific machine learning practical at engineering scale.
Projects
Indo-German Partnership Program
Hardware-aware numerical schemes and scalable algorithms for multi-node, multi-GPU supercomputers, with joint supervision, workshops and schools for young researchers.
AI-augmented wind-farm simulation
Real-time, environment-aware simulation of a wind-turbine farm combining deep learning, model order reduction and high-performance computing. Industry partner: Shell India.
Open-source software
Tensor-based Python library for solving partial differential equations with hp-variational physics-informed neural networks.
pip install fastvpinnsParMoon
Modernised program package based on mapped finite elements, with an object-oriented parallel design.
SParSH-AMG
Hybrid CPU-GPU algebraic multigrid and preconditioned iterative methods.
Publications
Books, journal articles, chapters and preprints. Select a period to filter.
About
Biography, roles and public work. Bios can be copied for introductions and press.
Roles
- Founder, ZenteiQ AiTech Innovations
- Professor, Department of Computational and Data Sciences, IIScOn academic leave from 2026
- Programme Director, IISc advanced programmes in Computational Data Science and AI, AI and MLOps, and Agentic and Generative AISince 2020
- Chairman, Department of Computational and Data Sciences2018–2024
- Co-Convenor, Kotak-IISc AI-ML Centre
- Founding Chair, M.Tech (Online) in Data Science and Business Analytics
- Adviser on AI integration for aerospace, defence and manufacturing
Academic background
- Alexander von Humboldt Postdoctoral Fellow, Weierstrass Institute for Applied Analysis and Stochastics, Berlin
- Research Associate, Department of Aeronautics, Imperial College London
- Dr. rer. nat. in Computational Mathematics, Otto-von-Guericke University, Germany
Public impact
In 2020 and 2021 the group built spatio-temporal and ensemble models of COVID-19 spread for Karnataka and India, to support vulnerability assessment and policy interventions. The projections were widely reported.
- 3 Apr 2021Times of India: second-wave projection
- 23 Oct 2020Frontline: Karnataka projection
- 1 Sep 2020Times of India: national projection
- 29 Aug 2020The New Indian Express: Karnataka peak
- 19 Jul 2020The Sunday Guardian: Karnataka projection
- 16 Jul 2020Times of India: IISc study projection
- 14 Jul 2020Firstpost: pandemic model
- 14 Jul 2020Deccan Herald: mathematical model
- VideoCOVID-19 discussion
- VideoCOVID-19 discussion
Teaching
Courses at IISc and for working professionals.
On academic leave from 2026
Generative and Agentic AI in Practice
A hands-on course on engineering and deploying generative and agentic AI systems: LLMs, agent frameworks, retrieval-augmented generation and fine-tuning, with a technical deep dive into the Transformer architecture and the Model Context Protocol. A capstone project ties it together.
| Course | Code | Term | Format |
|---|---|---|---|
| AI and Machine Learning | CCE | Aug–Dec 2024 | Online, evenings, for professionals |
| Introduction to Computing for AI and Machine Learning | DA 203o | Jan–Apr 2023 | Core course, M.Tech (Online) |
| Introduction to Computing for AI and Machine Learning | DS 226 | Aug 2022 | IISc course |
| High-Performance Computing and Data Science | CCE | Aug 2021 | Online, evenings, for professionals |
| Numerical Methods / Introduction to Scientific Computing | DS 288 / UE 201 | Oct 2020 | IISc course |
| Finite Elements: Theory and Algorithms | DS 291 | Oct 2020 | IISc course |
Contact
Academic enquiries here. ZenteiQ enquiries go to zenteiq.ai.
- Office
- Department of Computational and Data Sciences, Indian Institute of Science, Bengaluru, Karnataka 560012, India
- Profiles
- Google Scholar LinkedIn GitHub X Mathematics Genealogy