Ming-Kai Liu
Hi there, I'm Ming Kai Li, a Master of Science candidate in Computer Science and Engineering at UCSD with a strong focus on LLM systems and Computer Vision.
About
Hi there, I’m Ming Kai Li, a Master of Science candidate in Computer Science and Engineering at UCSD with a strong focus on LLM systems and Computer Vision. I enjoy turning research into practical AI systems, from LLM pipelines to knowledge graphs and clinical decision support.\n\nI have worked as an AI Engineer at Citrus Oncology and as a research intern in Academia Sinica and NTU’s Vision & Learning Lab, combining theory and hands-on deployment to help doctors and researchers make better decisions. I’m excited to apply my skills to challenging problems in data-driven software and AI.
- 1 year's experience
- Joined October 2025
Skills· 4
Languages
- ChineseFluent
- EnglishFluent
Industry experience
- Computers & Electronics
- Software & Internet
- Education
- Professional Services
- Media & Entertainment
More experts in San Diego, United States
Work experience
Engineered an end-to-end LLM pipeline to automatically extract key clinical data from unformatted image/text-based PDF Electronic Health Records (EHRs), creating a knowledge base to support a clinical chatbot. Designed and implemented a Retrieval-Augmented Generation (RAG) system to automate clinical follow-up report generation, grounding differential diagnoses and clinical reasoning in SOAP-formatted notes. Built a real-time conversational AI assistant that updates clinical reports as new information becomes available, improving physician workflow.
Extracted knowledge graphs from Cyber Threat Intelligence (CTI) reports to identify malware attack patterns. Enhanced entity extraction robustness by combining Large Language Models (LLMs) with Semantic Role Labeling (SRL), improving recall by 10% through more accurate detection of entities and their semantic relationships. Constructed a knowledge graph database and applied graph matching algorithms to identify potential Mitre ATT&CK TTPs, achieving a 44% recall rate.
Explored how diffusion model internal representations improve vision tasks like semantic segmentation. Researched the unlearning of diffusion models, specifically utilizing Textual Inversion to bypass the safety mechanism of Text-to-Image diffusion models.
Education
Hire Ming-Kai Liu today
Post a job and invite Ming-Kai, or contact them directly. No hidden commission.



