Demonstration of the Null framework, a multi-agent, multi-state scambaiting AI. This is a VERY early work in development and more to document advances of current vs future versions. With the eventual goal of being the first recorded instance of ‘self-driving scambaiting’ where a bot identifies a scammer, initiates a call, lets them access a virtual machine and perform a ‘diagnosis’ without any outside human control.
The bot works by calling, listening, processing the response, running through a list of known phrases to respond and picking the most appropriate one, seeing if an audio response is cached otherwise using a tts engine to dynamically generate the response, and then running some post processing effects to make it sound more ‘phone like’. The only external service being used is Google TTS, but it automatically caches every response. Once the responses are cached, the bot can run 100% locally on a single computer and only needs a SIP trunk. The bot can use any number of different voices, names, addresses, and phone numbers. Once the call connects, everything is hands off.