Dr. Pelgrino's DEFCON Presentation on Local Knowledge Graphs for OSINT

The conference presented at the DEFCON Conference by Dr. Pelgrino addresses the construction of local knowledge graphs to bypass rate limits and maintain opacity in open-source intelligence (OSINT) investigations. The topic explores the use of knowledge graph technology combined with the Resource Description Framework (RDF) to create offline queryable repositories.

Dr. Pelgrino, an information science expert with 20 years of experience for organizations such as the Pentagon, the DoD, and Homeland Security, presents a practical methodology. This includes data collection using tools like Playwright and Chromium to extract dynamic content (JavaScript, iframes), and then converting it into RDF for structured analysis. A case study analyzes the participants of Recon Village 2024 by cross-referencing data from Wikipedia and manual sources, aligned with the Friend of a Friend (FOAF) ontology.

Technical challenges include managing rate limits, query confidentiality, and the integrity of extracted data. The proposed solutions rely on local caches, language models (LLM) run locally (via Llama CPP, LM Studio), and the use of Graph Retrieval-Augmented Generation (GRAG) to optimize queries. A GitHub repository in Rust accompanies the demonstration, including a custom crawler and examples of SPARQL queries.

The conclusions highlight the importance of combining knowledge graphs with LLM for reproducible and scalable analysis, while controlling biases and limitations of external models. https://www.youtube.com/watch?v=yIUfntW_TxY