WHAT: TWed Talk: "ARGOS: Query Realignment for Secure LLM-to-Database Access"
WHEN: 4p, Weds, 29 Oct
WHERE: Winslow 1140
WEBEX: https://rensselaer.webex.com/meet/erickj4
VIDEO: TBD
EVENT PAGE: https://bit.ly/4o7qbUc
Please join us as Nipun Pathirage presents what promises to be an interesting TWed Talk, "ARGOS: Query Realignment for Secure LLM-to-Database Access." Pizza arrives approx. 3:30p, the talk begins at 4p.
DESCRIPTION: Large Language Models (LLMs) can generate insecure SQL queries, posing a significant risk to database security. We introduce ARGOS, an ontology-driven framework that automatically enforces access control. ARGOS acts as a secure middle layer, using a knowledge graph of policies to deterministically realign LLM-generated queries to be safe and compliant before they reach the database. This approach achieves 100% policy adherence, providing a robust and explainable bridge between LLMs and sensitive structured data.
BIO: Nipun D. Pathirage is a second-year PhD student at Rensselaer Polytechnic Institute (RPI). His recent research focused on evaluating Large Language Models (LLMs) using semantic technologies, enhancing their privacy, and preparing them for industrial applications. He is also venturing into the area of Knowledge Graph (KG)-based question answering with LLMs, specifically focusing on bridging lexical gaps and developing methods to create semantically-rich KGs from LLMs automatically. He holds a B.Sc. in Computer Science and Engineering from the University of Moratuwa and is currently advised on his doctoral research by Prof. Deborah McGuinness, Prof. Oshani Senevitne, and TWC researchers.
Pizzas and salads will be delivered at approx. 3:30p; the talk will begin at 4p.