From robot perception to AI safety.
I started in robotic perception: RL, computer vision, edge inference on real hardware. These days I work on a related problem, making AI systems safe and auditable.
MS Robotics at WPI (2023-2025), where I worked on deep RL for viewpoint planning in robotic metal scrap cutting, advised by Prof. Berk Calli. Before that, autonomous vehicle perception at ARTPARK/IISc and embedded systems at Bosch. Currently at GLACIS Technologies in Seattle.
Work
GLACIS Technologies current
glacis.io ↗AI Engineer. Applied ML research and software development.
Reinforcement learning based next best view planning
MER Lab ↗Deep RL for next-best-view: training a robot to figure out where to look to build a 3D model of unknown metal scraps for torch cutting. Point cloud processing and attention mechanisms.
Dynamic conveyor grasping
GitHub ↗Dual-camera robotic pick-and-place from moving conveyor belts. YOLO detection, learned grasp synthesis, closed-loop control. Handles 9cm position errors at 0.04 m/s.
Edge perception for warehouses
GitHub ↗Pallet detection and floor segmentation on Jetson. YOLO v11 trained on 15K+ images, 76.4% mAP, ~3ms inference via TensorRT FP16.
Healthcare AI evaluation
Built evaluation pipelines for healthcare AI at PsychedAboutAI: medical knowledge graphs, LLM-as-Judge scoring, automated red-teaming with DSPy.