For our agent observability, which is our productised version of what you are building, we upload all the data after the session closes as this was easier to implement in the first release. Showing stats in real-time will always be a better user experience, but for our initial release it was simpler to upload the stats on session close, then tackle real-time in a future release. (Should have also said, I work for LK devrel)

All feedback is appreciated 🙂 For telephony, the philosophy was that we wanted to make the onboarding easier with fully managed phone numbers and SIP provisioning: https://blog.livekit.io/introducing-livekit-phone-numbers-zero-to-ringing-in-60-seconds/, launched the same day as agent builder. In the livestream of the product announcement, the team created an agent with builder, then called into it with a number provisioned using LK phone numbers: https://www.youtube.com/live/Ly8_qOgPoBk?si=piOuv4Zu8IxJBpSX

🚀🚀 We'll have lots of cool announcements related to LiveKit, if you can't make it in person we're live-streaming the keynote on YT and I'll be hosting a "watch party" (for want of a better term) on our own community slack 🚀🚀

@Khizar Hayat you might be interested in our form agent that guides the user through a series of interview questions and stores the responses in UserData: https://github.com/livekit-examples/python-agents-examples/tree/main/complex-agents/nova-sonic (it's written for nova-sonic, but the principles apply to any LLM. That demo is 5 months old however, the more recent approach would be AgentTasks: https://docs.livekit.io/agents/build/workflows/#tasks (though this feature is still experimental). Although not yet GA, we also have a survey example using AgentTasks that I found impressive when I played with it last week: https://github.com/livekit/agents/blob/theo/v1.3/examples/survey/survey_agent.py

🙌 🙌