
The ScamTrap AI Voice Agent is a real-time counter-fraud system built to engage malicious callers, mitigate social engineering attacks, and gather actionable intelligence. By orchestrating a finite-state machine with streaming speech-to-text, low-latency language models, and text-to-speech pipelines, the agent adopts realistic counter-personas (such as a confused senior or hesitant target) to deliberately prolong fraudulent phone interactions. As the caller speaks, the agent streams live audio to transcribe speech in real time, uses stateful conversational pacing to introduce realistic stall tactics (mishearing details, fumbling for items, feigning tech confusion), and responds with synthesized voice audio. Operating completely out-of-band from the speech loop, an asynchronous intelligence extraction engine parses conversational transcripts to extract critical Indicators of Compromise (IOCs)—including claimed identities, mule bank accounts, malicious URLs, callback phone numbers, urgency vectors, and social engineering scripts—instantly rendering them onto a live threat telemetry dashboard.
30 Sep 2026