I am employed as a Software Engineer. I work closely with other developers and the project manager to design software solutions that meet customer requirements. My responsibilities include building general services and quickly identifying and resolving technical issues. I also focus on identifying performance bottlenecks and integrating our systems with other tools to ensure efficient operation.

Led and participated in the establishment of the medical fraud detection system.

Led to building an LLM-assisted medical record information extraction system by designing prompts to extract key information. Focusing on the quantity and price of medical services, and the frequency of patients who receive these medical services, along with the doctor’s ID who provided them. The data will be fed to the decision process based on the rule base generated. This enables medical experts to extract crucial information from various file types and make swift decisions to determine if it warrants further investigation.

Led to establish the decision-making mechanism based on disease–drug relationship knowledge graph, which was implemented with Neo4j. With the effort of extracting frequent pairs from large-scale historical medical record data using automation scripts, and collaborating with medical experts to validate and refine the relationships of disease and drug, supporting downstream use in detecting overprescription and other fraudulent behavior. Stored these relationships and behaviors in the rule base, and accelerated the fraud recognition speed for future cases, which reduces the number of human decisions.

Built a detailed terms classification system with several trained XGBoost models. Extracted key information from poor-quality medical records and formatted classified key details in a fixed structure, acting as an assistant to build the knowledge graph system. This saves more than 90% of the time and effort that the medical expert pays to obtain key components from these medical records.