Developing a diffusion policy-based learning-from-demonstration pipeline enabling bimanual KUKA iiwa to autonomously perform complex engine inspection tasks
Developing data augmentation pipelines for trajectory and vision data to boost sim-to-real transfer and improve policy robustness from expert demonstrations
Architected a modular ROS2 framework for seamless communication between low-level robot interfaces and high-level application logic
Leveraged PrefPaint and inpainting diffusion models to generate photo-realistic, context-aware synthetic frames for domain randomization and improved model generalization
Designed and trained custom classification networks, achieving 93.6% accuracy to boost vehicle detection in sensor-limited environments
Developed graph-based representations of GPS probe data to model truck dwell patterns and trip dynamics, leveraging spatiotemporal feature engineering and road network topology enabling predictive freight analytics for better driving experience
Developed 3D scanning techniques to create CAD models of bump caps from various manufacturers.
Designed a custom classification model to detect bump cap types and anchor points, facilitating seamless integration of our product with any manufacturer.
I even had the opportunity to explore management side of things - talking to various manufacturers and distributors for potential collaborations.
Additionally, I collaborated with Kethan Gudi on designing custom electric boards to meet our design requirements and assisted Dylan Weiss with her viral testing experiments. These experiences broadened my expertise across multiple disciplines, including computer vision, mechanical design, circuit design, and biology—all of which I found deeply rewarding
✨ During my co-op, I was featured in an article by The Michigan Daily highlighting Taza Aya's innovative approach to respiratory protection. The article discusses how the company developed an invisible air curtain that filters more than 99% of aerosols in the user's breathing area, utilizing a technology similar to a visor, with blown air replacing the glass. ✨
👉🏼 You can read the full article here - UMich startup builds mask that filters more than 99% of aerosols
After completing my college, I joined Brillio as a Data Engineer. I experienced what corporate life really is. Met new people almost everyday.
As a fresher, I was put into training on the big data technologies such as SQL, HQL, Hive, HDFS, Map-Reduce concepts, Spark, Pyspark, Python, Amazon Web Services, for about 3 months. After completing my training period, I was part of 'e-Bay DCP Live' project.
Being part of the data team of the project my main responsibilities included maintaining and monitoring the Airflow DAGs, analyze the data scraping process, and reduce the cost incurred by the external vendors by at least 30%. Implemented a sound and an efficient way to reduce the cost, and maintain the data availability, without compromising the data quality. This data was used by different teams as well.
Recipient of Customer Value Success Award - FY 2022
After my sophomore year, I worked under Dr. Raghunandan, Associate Professor at Mechatronics department, MIT, Manipal. My work gravitated towards Modular Robots, more precisely, I designed and modeled individual modules of a serial link manipulator, using Catia v5, and assembled them and evaluated the linkage parameters using the Denavit-Hartenberg (D-H) algorithm. Also built a dynamic response model using MATLAB, to actively change the link parameters, such as link length, joint angles, and the prismatic distance of the last link of the serial manipulator or the robotic arm. Click here to see the project report
A 7 day summer visit, in CNC India , Tools & Services (P) Limited , Peenya , Bengaluru, which involved exposure to robotic machinery used in manufacture of brake shoes of executive cars and buses, fixtures for manufacturing chassis of two wheeler's, and silencer design of the same.