Aman Pawar

Thesis: Probabilistic Causal Inference in Complex Dynamical Systems. Advisors: Prof. Peter Benner (MPI) and Prof. Philipp Hennig (University of Tübingen). Focus: scientific ML, Bayesian inference, operator learning, probabilistic numerics, uncertainty quantification, causal discovery.

CGPA 9.4/10. Thesis: “Electronic Health Records for Individuals with Diabetes is All You Need.” Coursework: Reinforcement Learning, Mathematical Foundations for ML, Systems Biology, Advanced Robotics.

Selected participant, advanced coursework in representation learning, reinforcement learning, causal inference, foundation models, and ML for healthcare.

CGPA 8.33/10. Mathematical foundations: finite element analysis, numerical methods, differential equations, optimization, mechanics.

Co-developed and validated “Arogya Sankalp”, an AI oral-cancer screening app for primary health centres; drafted a clinical cluster-RCT and funding-ready proposal for the Evidence-for-AI-in-Health initiative with ICMR, IISc, and the Ministry of Ayush.

Authored a 25-page AI product & hardware–software specification for Krutrim’s custom AI silicon, vehicle perception, and MoveOS 6; led a cross-functional squad of 10 across AI/ML, embedded software, and hardware. Inference/quantization design cut projected on-device latency ~2.7× on benchmark workloads.

Built prototype modules for automated data-stream orchestration and adaptive model quantization; contributed to multiple IP disclosures on automated pipeline synthesis and field-trainable inference.

Graph-based causal-discovery pipelines on longitudinal EHRs (MIMIC-IV, ~70k records), ~18% improvement in structural-recovery F1 vs classical temporal-causal baselines; uncertainty-aware pipelines for diagnostics under noisy/incomplete data.

Physics-informed and Hamiltonian Graph Neural Networks for physically consistent rigid-body dynamics (<3% long-horizon energy drift vs >15% for unconstrained baselines); differentiable ML-driven topology optimization for truss design.