Building ML systems that work in practice: From predictive maintenance pipelines to reinforcement learning agents that beat human baselines.
I'm an AI/ML engineer with a Master's in Artificial Intelligence from Northeastern University, with a background in Information Technology and Cybersecurity from University of Mumbai.
My work ranges from building predictive maintenance models that cut elevator downtime by 26%, to training deep RL agents that reduce task completion from 56 steps down to 9. I like working on problems where the numbers actually move.
A strong theoretical foundation combined with hands-on research experience
Industry internships and technical community leadership
Selected projects across ML, computer vision, and data engineering
Lightweight F1 race winner prediction model using FastF1 and LightGBM. Trains on historical race data to predict win probabilities based on qualifying results, driver form, and team performance.
Features rolling averages, circuit-specific modeling, and team performance metrics for accurate race outcome predictions.
A privacy-focused personal finance app that turns raw transaction data into mood-adaptive "Money Narratives" — spending insights without storing any sensitive user data.
Includes an LLM-powered assistant ("Finn") via the Gemini API for contextual follow-up queries on anonymized stats. Built following Shneiderman's Eight Golden Rules for user control and transparency.
ML-powered lung cancer detection system using Convolutional Neural Networks to analyze CT scan images. Enables early detection when treatment is most likely to be successful.
Features advanced image processing techniques including edge detection, segmentation, and feature extraction for accurate medical diagnosis assistance.
CNN-DQN agent trained across 1 million episodes in a dynamic simulation with moving obstacles and adaptive rewards. Achieved a 35% boost in navigation efficiency.
Uses experience replay, target networks, and hyperparameter tuning to achieve stable convergence over Q-learning baseline.
Algorithmic music generation system integrating rule-based harmony modeling with variational autoencoders to replicate classical piano textures.
Features tonal extraction, chord progression modeling, and latent-space regularization inspired by Mozart, Bach, and Beethoven.
Tools and frameworks I use regularly
PyTorch, TensorFlow, Scikit-learn, OpenAI Gym. Reinforcement learning (DQN, Q-Learning), deep learning (CNNs, VAEs, RNNs), NLP, computer vision, and model training with feature engineering.
Linux (Ubuntu), Windows Server, macOS. Docker, VirtualBox/VMware, AWS, server configuration. Familiar with Agile methodology, network protocols, and hardware troubleshooting.
Git, GitHub, Bash, PowerShell, Python scripting. CI/CD pipelines, DevOps workflows. Database work with MongoDB and MySQL. Cloud deployment on AWS.