Research focus on human-in-the-loop perception models in bite-acquisition robotics system focusing on what to query the user

• LLM-based Avatar Interface: Collaborated with a team of 5 to build a real-time, multimodal medical screening tool leveraging voice, vision, and text interactions. Deployed AWS infrastructure for secure voice data storage and management; integrated OpenAI API for NLP capabilities; implemented Speech-to-Speech functionality using AWS services and React transcribe tools. Established WebSocket connections for low-latency, real-time streaming and automated CI/CD workflows with Jenkins, Docker, and Git.• Voice Agent App: Created a voice agent app utilizing OpenAI GPT models and AWS for Speech-to-Speech processing. Integrated Google Maps and Google Search APIs for live information retrieval, deployed for phone screening in settings like hospitals and restaurants.• Optimized System Performance: Enhanced API latency by 25% through parallel processing techniques, leveraging async/await, Promise.all, and worker threads to improve system efficiency and waiting time.• Llama-3-based Chatbot: Fine-tuned the open-sourced Llama-3 large language model (LLM) using a medical text corpus for generating medical reports. Achieved a 20% improvement in report quality, as measured by the ROUGE metric, compared to a T5 model baseline.• Brain Computer Interface: Architected full-stack EEG analysis platform using React and FastAPI. Im- plemented RESTful API endpoints for data processing with MNE library; designed interactive visualization dashboard; containerized application with Docker and established automated testing pipeline using Jenkins.

- Advised by Nikolaos Bouklas- Researched LLM-based symbolic regression on ordinary differential equations. The work built from the ICLR paper SNIP: A Multimodal Symbolic-Numeric Pretraining for Math.

- Advised by David Goldberg- Research in data inaccurate detection- Lead a team of group of five. Organize weekly meetings and push the agenda to meet the project goals

- Advised by David Goldberg- Research in RPCC queuing time reduction

Research with Prof Weiqing Gu on applying ML-EEG for early diagnosis of mTBI and the ML-EEG market.