resume
Education
Carnegie Mellon University Expected Dec 2026
Master of Science in Machine Learning Pittsburgh, PA
Anna University, College of Engineering, Guindy June 2022
B.E. (Honors) in Computer Science & Engineering Chennai, India
Work Experience
Google Hyderabad, India
Software Engineer III Nov 2024 – Jul 2025
Software Engineer II Jul 2022 – Oct 2024
- Drove risk strategy design for Google Wallet via PIX in Brazil, developing mitigation frameworks to reduce transaction losses across P2P and P2M payment flows for the August 2024 launch.
- Designed and deployed feature engineering pipelines in Java and rule-based decision policies in Python integrating ML model outputs for real-time transaction risk assessment across Google Pay US and Wallet Brazil.
- Owned design, implementation, and validation of payment risk infrastructure extensions for recipients external to Google, including entity representation, control propagation, and downstream enforcement across the transaction pipeline.
- Engineered ML pipelines for delinquency fraud detection on Stored Value transactions, spanning data simulation, GBDT model training in TensorFlow and offline evaluation against production baselines.
- Built real-time monitoring dashboards and operational runbooks for regulatory compliance in Brazil. Collaborated with Trust & Safety and Product teams on post-launch metrics, thresholds and customer impact.
Research Experience
Forge Lab (Prof. Virginia Smith), Carnegie Mellon University May 2026 – Present
Graduate Student Researcher (LLM Safety and Alignment) Pittsburgh, PA
- Developing post-training interventions for LLM alignment, focused on identifying and suppressing unsafe persona-driven behaviours through on-policy consistency training and RL.
- Conducting a user study for Safety Nudges, a real-time Chrome extension that audits chatbot conversations and surfaces contextual warnings for overconfidence, sycophancy, anthropomorphization and unsafe responses.
Human Sensing Lab, Carnegie Mellon University (Prof. Fernando De la Torre) Oct 2025 – Present
Graduate Student Researcher (Multimodal Models) Pittsburgh, PA
- Developed geometry-guided visual token pruning methods for multi-view 3D Visual Question Answering with 2D Vision-Language models, reducing inference cost by 60%, FLOPs by 88%, and KV-cache usage by 86% while maintaining comparable benchmark accuracy (in collaboration with Meta Reality Labs). Preprint coming soon.
- Working with Vision-Language Action models (in collaboration with Fujitsu Research).
Indian Institute of Technology (IIT), Kharagpur May 2022 – Sep 2022
Deep Learning Intern Remote (Kharagpur, India)
- Trained segmentation models in PyTorch for depth estimation from simulated colonoscopy images as part of the KLIV group and studied effective post-processing techniques. Presented at the MICCAI 2022 Endoscopic Vision Challenge.
Projects
Watch & Learn: Teacher-Student Distillation for Robotic Arm Tasks (GitHub) Nov 2025 – Dec 2025
- Developed a vision-only student agent from distillation of a full-state teacher policy for camera-based robotic manipulation on MetaWorld tasks. Achieved 90% of full-state teacher performance via teacher-guided RL with imitation learning.
Environmental ML: Modelling Sequential Disaster Cascades (GitHub) Jan 2026 – Apr 2026
- Built an end-to-end disaster cascade prediction pipeline on 15 years of NOAA Storm Events data, using multilabel classification, weather embedded neural networks and graph neural networks to model rare secondary hazards and spatial cascade patterns.
Adaptive Day Planner: LLM-Based Task Scheduling and Voice Agent (GitHub) Feb 2026
- Built an AI-powered adaptive planner using Mistral LLMs, FastAPI and Eleven Labs speech APIs to dynamically schedule tasks, replan sessions based on user feedback and support focus sessions via conversational body doubling. (Mistral AI Worldwide Hackathon, SF)
Skills
Languages Python, Java, C/C++, SQL, JavaScript ML / Deep Learning PyTorch, TensorFlow, Keras, NumPy, Pandas, scikit-learn, Hugging Face, OpenCV, Gymnasium Developer Tools Weights & Biases, Docker, Git, FastAPI Research Interests VLMs, LLMs, post-training, RL, multimodal reasoning, technical AI safety