• Contributed on the convex mesh data pipeline for BEHAVIOR-1K, a benchmark for evaluating Embodied AI solutions through interactive household activities with physics simulations supervised by Dr. Fei-Fei Li.

• Developed teleoperation features for OmniGibson, Stanford-developed robot/environment simulation platform built on top of NVIDIA’s IsaacSim, including operating the robot simulation with Nintendo Switch Joy-Cons and an Adafruit Gamepad with VR headset.

• Scripting in Autodesk 3ds Max, integrating CoACD performance metrics, where achieved up to 39% speed-up in internal sanity check scripts for integrity checks of physics simulations.

• Implemented Named Entity Recognition (NER) language models for low-resource languages Kazakh (95% F1-score) and Uyghur (77% F1-score) utilizing syntactic similarities across high-resource languages like Turkish through transfer learning (supervised by Dr. John Bauer and Dr. Chris Manning).

• Enhanced the Stanford-developed Stanza natural language analysis package, with an 11% increase in parsing speed for nested NER tags, and added generalized transliteration across scripts and word span correction features.