AI Researcher | Leader

Jay Thiagarajan

Principal Research Scientist at Amazon AGI leading frontier model customization across diverse industry verticals.

PhD from Arizona State University. Previously at Apple and Lawrence Livermore National Labs. Track record of shipping high-impact AI solutions across multiple industry verticals.

Funding & Leadership
🚀 Lead investigator of projects funded by DOE, DARPA, DNN (~$15M)
🚀 Holder of 15+ patents on deep learning and AI applications

Awards & Recognition
🏆 Recipient of LLNL/DoE Early Career Recognition Award
🏆 Recipient of WCI Gold Award for leadership in COVID-19 modeling
🏆 Best paper awards at premier conferences/journals

Service & Community
🤝 Member of the ML Commons team on Medperf development
🤝 Invitee at Dagstuhl symposia, BIRS workshops, STS forum
🤝 Advisor at Doctoral Consortiums, NSF MSGI program, DOE CSGF

Research Areas: Multimodal foundation models, visual generation, computer vision, uncertainty quantification, inverse problems, explainable AI, robust/safe AI, AI4Science, AI4Health.

Career

Pr. Research
Scientist
Amazon AGI

Staff Research Engineer
Apple

Research
Lead
LLNL

Strong research portfolio (300+ publications, h-index: 40) with high-impact delivery in consumer products, sciences, healthcare, and security.

Managed research teams (size: 10-15) and mentored post-doctoral researchers, and staff scientists/engineers.

Influenced research directions, built technical propositions for strategic investments, and delivered products.

What’s new?

New paper!

Revisiting NeRN: Optimizing Training Strategies for Weight Reconstruction

We improve INR-based weight parameterization (NeRN) with new training strategies that achieve high-fidelity weight recovery and better generalization.

ICML’26 Workshop on Connecting Low-rank Representations in AI

Preprint on VLA training

Learn to Judge, Then Learn to Act: Closing the Physical Reasoning Gap via Rank Prediction Pre-Training

Teach VLMs to rank grasps before predicting actions and they learn physical reasoning that transfers to better robotic manipulation.

Preprint coming soon!

AWS re:invent 2025

Amazon Nova Forge: Build your own frontier models using Nova

Nova Forge customers can start their development from early model checkpoints, blend proprietary data with Amazon-curated training data, and host their custom models securely on AWS.

New product launched at Amazon re:invent 2025

A Peek into my Research

Multimodal AI

Developing and leveraging multimodal LLMs and representation learners for applications in vision, sciences and embodied AI

Learn More…

Generative Models

Designing generative models and innovating the use of generative priors for inverse problems and experimental data calibration

Learn More…

Safety & Robustness

Advances in model generalization, test-time scaling, uncertainty quantification and failure characterization

Learn More…