• Researched open/closed-source LLMs & SLMs for function calling on resource-constrained edge devices; optimized latency and token utilization.

• Enhanced Copilot agent performance through prompt engineering (CoT, RAG), vector database design, and task-specific fine-tuning.

• Led cross-functional team to benchmark hologram accuracy using spatial anchors (multi-anchor, object anchor); established QA pipelines and performance baselines.

• Worked from the early stage to the shipping stage for Windows Subsystem for Android (WSA), a system service that enables Windows OS to run Android applications (.apk)

• Bridged AOSP to Windows OS notification using C++, Java, proto buffers

• Conducted Sparse-to-Dense Optical Flow research using deep learning under the supervision of Dr. Chakrabarti

• Using Tensorflow to build UNet, ResNet, Encoder-Decoder, and other custom neural network architectures

• Using Tensorflow, implemented a ResNet-50 (2015), trained, and attained an accuracy of 97% on LFW dataset

• Built a face recognition system that is added to the Human Perception Package SDK