Working with Dr. Thomas Ploetz and serving a class size of ~300 students, specializing in Optimization and Machine Learning.

Working with Dr. Thad Starner for CS 6601/3600 Artificial Intelligence. Serving a class size of 1800 students and specializing in Bayesian networks, machine learning and non-convex optimization.

Designed a large-scale menu item recommendation system serving a customer base of 15.3M individuals based on temporally weighted taste and ingredient embeddings.Designed and trained deep Markovian models for predicting customer phase transitions in terms of visit patterns and spending habits. Characterized long-term customer dynamics and drove never-seen-before business insights generating a revenue potential of $0.6M/quarter. Built a promotional target quantifier and budget allocator based on phase transition probabilities. Also enabled incrementality measurement for customer inducements.Engineered Ripple, an end-to-end data transformation tool, reducing cloud costs and enabling concurrent queries and an interoperable Spark-based workflow between Palantir Foundry and local compute.Built a low-cost, scalable computer vision system called Occuvision, for dispatch window occupancy monitoring. Designed a fine-tuned YOLO based detector and tracking system robust against reflections and occlusions. Built an intuitive desktop application for the ground staff to monitor and visualize pickup times on a per-plate basis. Also designed a customizable alert system using Twilio API to notify the staff of service events like high urgency, high arrival rate, low service rate and pickup time violations. Occuvision enabled granular operational data generation for inventory and staffing prediction.

Areas of work: 1. Visual odometry2. Simulation of autonomous ground and underwater vehicles3. Vehicle dynamics modelling4. Model predictive control5. Sensor fusion and localization6. Pedestrian simulationCharacterized numerical models and drift for sensors like the IMU, LIDAR and GPS and studied localization performance in GPS denied environments and during transitions between GPS equipped and denied environments. Developed a novel fast nonlinear model predictive controller based on self-similar sampling; improving control frequencies from 20.5x up to ~40Hz. Gained significant experience with model discovery and dynamics.Authored proprietary plugins for Gazebo and Rviz, to simulate and control adverse environmental conditions at runtime. Designed a graphical pedestrian swarm specification plugin, with Rviz interactive markers.

For Dr. Pravin Singru (Machine Design and Drawing)

For Dr. Sachin Waigaonkar (Production Techniques II)

Established forward and inverse, kinematics and dynamics for the SSR 3-DOF dual manipulator.Developed a novel reactionless control algorithm with situational infeasibility proofs, for the robotic arms in micro-gravity environments. Proposed a generalized joint-space trajectory generation method with explicitly controllable maximum angular rates. SSR to be launched to low earth orbit with PS-4 stage of PSLV. Garnered formal appreciation for my work by Dr. Sam Dayal Dev, Director (Inertial Systems Unit).

Responsible for the pitching and budgeting projects, monitoring their progress and handling funds.Projects I incubated during my tenure: 1. Recreating sonoluminescence2. Audio detection with Michelson interferometry3. Comet trail simulation.