Autonomous Video Hunter: AI Agents for Real-Time OSINT

Imagine discovering critical intelligence hidden inside live video streams faster than any human analyst could. We'll begin with a compelling hypothetical scenario: a breaking news livestream unintentionally captures crucial clues about a missing person's location, but overwhelmed human investigators miss the moment. Inspired by real world challenges investigators face daily, this scenario motivated us to build Autonomous Video Hunter (AVH), a system of AI powered agents that scour video content in real time to extract actionable OSINT.

Technical core:

We'll showcase how AVH combines open source AI models for image recognition and audio transcription, orchestrated by custom Python based agents. These agents autonomously analyze video streams, detect critical visuals, logos, speech keywords, and quickly cross reference these clues against online databases and OSINT repositories.

Live demo:

Experience AVH live as it identifies a target logo and relevant context (e.g., social media profiles and geolocation clues) from a random video clip in mere seconds. We'll also address practical challenges, from reducing false positives to scaling efficiently across multiple simultaneous streams.

By the end of this lightning talk, attendees will understand how autonomous agents transform overwhelming video data into OSINT insights rapidly and effectively. We'll also share a lightweight open source AVH tool for the OSINT community to use and build upon.