Announcement, slides, notebook. All resources can also be found in my archive. Did you miss the talk? Check out the write-up’s part 1 and part 2.
At ODSC West 2024 I talked about building modern and reliable voice bots using Pipecat, a recently released open source tool. I gave an overview of the general structure of voice bots, of the improvements their underlying tech recently saw, from the first Whisper release to GPT 4o Realtime, and the new challenges that developers face when implementing one of these systems.
The main highlight of the talk is the notebook where I implement first a simple Pipecat bot from scratch, and then I give an overview of how to blend intent detection and system prompt switching to improve our control of how LLM bots interact with users.