Developed Livestack, a Realtime AI App Framework, which can easily combine and implement any AI agents.

Key features:

• Implemented a Discord webcrawler bot that parses for all historical and real-time messages.
• Integrated the bot with speech app using the framework.
• Enabled GenAI (Claude) to summarize and update based on real-time speech and Discord conversations.
• Displayed results on meeting screens for easy access and collaboration.

Developed a comprehensive hospital large-screen display module for pathogen data analysis, management, and alert platform.

Developed an automated React App System for detecting senior mild cognitive impairment by analyzing gait.

Key features:

• Using mediapipe and openCV to analyze gait gesture and velocity of the filmed video.
• Implemented a camera calibration algorithm using a chessboard collection calibration with openCV and 3Dto2d remapping algorithm for image comparation; Allowing for easy and general use of any camera in common daily scenarios.

Developed a QMastery pipeline for enhancing LLMs performance using self-generated question augmentation to solve the problem of limited training data resources

Key features:

• Enabling self-generating 5k and more data based on 100 seed data, and enhanced models performance by fine-tuning on the generated data.
• Achieved 10% improvements in the fine-tuned LLM models' accuracy for SAT-Math, GSM8K (grade level math), and Aqua-rat (graduate level math) datasets, demonstrating the efficacy of the novel data augmentation strategy.