Embedded Files
A Local, Portable, and Secure Knowledge-Enhanced Agentic AI System for HIPAA Compliance (UNIVERSITY OF WEST FLORIDA)
Advisor: Guillermo A. Francia III
In my dissertation, I developed a local, secure, agentic system for verifying the compliance of medical artifacts. The system operates entirely offline on an M4 MacBook Air with 16 GB of RAM, without relying on third-party servers or an internet connection. It incorporates HIPAA compliance rules and the company’s confidential regulations into a FAISS index for efficient information retrieval. Once the dense vector database is built, users can assess documents for compliance.
In another words, when a new document is uploaded, it is converted into embeddings and passed to the retrieval node. This node queries the FAISS index to fetch the most relevant knowledge chunks. The retrieved information is then forwarded to the analysis node, where the system compares the document against stored HIPAA and organizational rules. The analysis node determines compliance status with supporting reasoning. Finally, if any Protected Health Information (PHI) is present, it is redacted before the response is sent to the local server and displayed on the web interface.
Additionally, I could be able to develop agentic AI systems using domain-specific datasets in Agricultural Accidents, Pipe Leak Detection, and Water Quality Monitoring. These systems significantly improve accuracy, reliability, and practical applicability. The final output of the system provides a clear compliance decision accompanied by detailed, human-readable reasoning — ensuring transparency, traceability, and trustworthiness in automated compliance assessment.
Page updated
Google Sites
Report abuse