May 1, 2025
Very happy to get to work with @FireworksAI_HQ on the Voice AI and Voice Agents course we're doing in May. Fireworks is giving every student in the course $500 in credits. ($7k total in free credits for every student!)
I first met Ray Thai from Fireworks at a hackathon more than a year ago, right after Fireworks released FireFunction, an innovative LLM optimized for function calling.
Ray walked me through how they trained the FireFunction model and why. For the hackathon, I built a voice AI agent that used function calling to step through a "workflow" of conversation states defined in a core prompt.
This approach of using function calls to navigate through conversation states has now become a standard way to script voice AI conversations.
Ray is going to hang out during the course, talk about models and inference, and host Fireworks office hours.
Here's the sign-up link for the course (with a $100 discount coupon attached to the link):
https://t.co/yjxxM3Brtw
Here are the @FireworksAI_HQ developer docs:
https://t.co/FuZHUqm9n7
If you're interested in how to use function calling effectively in voice AI agents, there are extensive notes about function calling reliability, latency, context management, and async function calls in the Voice AI Guide:
https://t.co/ee78el9SGW
There's a voice AI workflows state machine library built on top of @pipecat_ai, now, called Pipecat Flows. It's really cool (and completely open source). It even comes with a graphical workflow editor:
https://t.co/tRtSVtyrtr