Research Projects
Clinpax Medical AI Ambient Scribe
Healthcare AIA multi-agent AI Scribe architecture that generates clinical notes, billing codes, and letters from patient–doctor consultations, benchmarked against state-of-the-art open-source and commercial scribes on clinically vetted evaluation matrices. Includes a clinical Q&A co-pilot built on a hierarchical vector DB and evolving knowledge graphs of therapeutic guidelines for high-accuracy, low-latency results.
CEO-Bench: Business Profit Maximization Agent
Multi-Agent SystemsA multi-agent system designed for real-world business decision-making, benchmarked on CEO-Bench for maximizing profitability through coordinated long-horizon agentic reasoning.
Mathematical Proof Writer
Multi-Agent SystemsA multi-agent system that validates newly proposed mathematical lemmas and theorems, developed as part of research into accuracy maintenance for long-horizon agentic tasks.
GENIUS Education Platform
EdTechA learning management system with state-of-the-art sketch reasoning and live sketch monitoring/co-creation using a multi-agent framework.
Emotional and Persuasive Sales Assistant
Conversational AIA multi-persona, emotionally aware/reactive, and persuasive sales assistant (chatbot) for e-commerce platform such as Alibaba.
SEAL: Simultaneous Exploration And Localization
Multi-Robot SystemsDeveloped an integrated solution for simultaneous exploration and localization in multi-robot systems. This approach enables robots to navigate GPS-denied environments while simultaneously building maps and maintaining accurate relative positions.
CQLite: Communication-Efficient Multi-Robot Exploration
Reinforcement LearningDeveloped a communication-efficient multi-robot exploration system using coverage-biased distributed Q-Learning. This approach significantly reduces communication overhead while maintaining exploration efficiency in complex environments.
PhysicsAssistant
Educational RoboticsDeveloped an LLM-Powered Interactive Learning Robot for Physics Lab Investigations. This system integrates YOLO object detection, speech recognition, and large language models to create an interactive learning experience for students in physics laboratories.
G-SciEdBERT
Natural Language ProcessingDeveloped a contextualized language model for Science Assessment Tasks in German. This specialized model is fine-tuned for automatic scoring of German-language science education assessments, providing a valuable tool for educators and researchers.
DGORL: Distributed Graph Optimization for Localization
Graph AlgorithmsDesigned a distributed graph optimization-based relative localization system for multi-robot systems. This approach enables accurate positioning without GPS in dynamic environments, using only local communications between robots.
Fine-tuning LLMs for Automatic Scoring
Educational AIImplemented specialized fine-tuning techniques on large language models like ChatGPT for the automatic scoring of educational assessments. This work demonstrates significant improvements in scoring accuracy, outperforming traditional approaches.
Funded Research Projects
GenAI Empowered National Initiative for Upholding STEM+C Education (GENIUS)
2024 – 2029 Key Personnel Institute of Education Sciences, $10,000,000
Working on a large-scale initiative to integrate generative AI in STEM and computing education.
Automatic Measurement of Student Reasoning Patterns in Scientific Argumentation (Auto-SPIN-SA)
2024 – 2029 Key Personnel Institute of Education Sciences, $1,999,682
Developing AI systems to automatically measure and analyze student reasoning patterns in scientific argumentation tasks.
Supporting Instructional Decision Making: The Potential of An Automatically Scored Three-dimensional Assessment System (PASTA)
2023 – 2025 Postdoc Research Associate National Science Foundation, $2,999,996
Working on designing automatic assessment items, rubrics, and training machine learning models for automatic scoring in three-dimensional science education.