ML Researcher · Scientific Computing · Distributed AI Systems
B.Tech CS @ University of Delhi · B.S. Data Science @ IIT Madras
Research Intern @ Universidad de Chile · GSoC 2026 · GATE 2026
- Physics-Informed Neural Networks (PINNs) and PDE-constrained deep learning
- Stochastic differential equations for biological and physical system modeling
- Adversarially robust deep reinforcement learning
- Federated learning for privacy-preserving distributed ML
- Edge AI systems and distributed benchmark-driven inference
UC Chile — Research Intern (Jan–Aug 2026)
Building physics-informed neural surrogates for cancer cell migration modeling using coupled SDE-PDE frameworks. Implementing uncertainty quantification pipelines under stochastic biological constraints.
Adversarially Robust Federated Learning for UAV Networks
Developing a Byzantine-robust federated learning framework for UAV ad hoc networks, extending FLTrust with an adaptive trust threshold mechanism — replacing the static threshold used in prior state-of-the-art (FLTrust, CRFL) to improve robustness under dynamic adversarial conditions including jamming, blackhole, and Sybil attack vectors.
| Project | Description | Stack |
|---|---|---|
| PINNs for Cerebrovascular Hemodynamics | HFM framework encoding 3D Navier–Stokes as physics-residual loss; pressure/velocity inference from sparse vascular observations | PyTorch, NumPy, ParaView |
| SHAP-Attributed AQI Decomposition | Federated regression decomposing 2,500+ daily AQI observations; SHAP attribution for meteorological vs anthropogenic variance | PyTorch, scikit-learn, SHAP |
| LLM Automation Agent | Multi-step autonomous agent with tool-use, persistent memory, and branching decision logic | Python, LangChain |
ML/Science PINNs · DRL (DQN/PPO) · SDE Modeling · Federated Learning
XAI/SHAP · LLM Evaluation · Numerical PDEs
Frameworks PyTorch · Stable-Baselines3 · LangChain · OpenAI Gym · scikit-learn
Systems Git · GitHub Actions · LaTeX · ParaView · Flask
Languages Python · C++ · Java · SQL
- GSoC 2026 — Selected under ML4Sci organisation
- GATE 2026 — Qualified
- JEE Main — 94.8 percentile
Open to research collaborations in Scientific ML, edge AI systems, and distributed learning.