Projects

Agentic Driving Coach
  • Developed a reactor-MoC using Lingua Franca to enforce determinism in agentic AI-powered CPS.
  • Focused on determinism with human-in-the-loop, AI agents, and dynamic environments.
  • Designed a driving coach as a case study for safety-critical CPS.
11/2025 – Present
Personalizing LLM using RAG and KG
  • Developed a RAG framework using knowledge graphs to reduce hallucinations.
  • Integrated calendar, contacts, and location for personalization.
  • Improved response accuracy with minimal latency overhead.
07/2024 – 03/2025
Key Point Detection
  • Developed HRNet model using PyTorch with COCO dataset achieving ~99% accuracy.
  • Trained custom dataset achieving ~98% accuracy.
  • Converted model to TensorFlow Lite and deployed on Raspberry Pi.
  • Enabled real-time keypoint detection visualization.
10/2023 – 12/2023
Enhancing Large Language Models
  • Improved QA performance using BERT with PyTorch.
  • Enhanced LLaMA and T5 using transfer learning.
  • Evaluated improvements using ROUGE metrics.
09/2023 – 12/2023
Learning Complex Dexterous Manipulation
  • Applied DAPG and reinforcement learning for object relocation tasks.
  • Implemented behavioral cloning and natural gradient methods.
  • Improved success rate from 98% to 99% using PPO.
01/2023 – 05/2023
Style GANs Translation
  • Generated images from sketches using GAN-based translation.
  • Addressed limited dataset issues with image-to-image techniques.
  • Reduced training time by 30% using Google Colab GPUs.
08/2022 – 12/2022