Build a browser-based voice agent that takes bookings, then trace, debug and verify it when the call goes wrong. One appointment-booking case runs across both days, through interruptions, callers who change their minds, slow responses and failing back-end systems.
Third or fourth week of January 2027IIT Hyderabad, 10 AM to 5 PM IST each dayIn person or online, facilitated by Dr. Praveen Tammana and Koundinya N. V. S. S.
New dates, same hands-on experience!
Engineering Real-Time Voice Agents is moving to the third or fourth week of January 2027. Thank you for your flexibility. We look forward to building with you!
No payment is taken. Expressing interest is not a confirmed registration.
Developers comfortable with Python, terminal commands, virtual environments and REST APIs. Familiarity with async Python helps.
02
What to bring
A laptop and a wired headset with a microphone; the workshop uses browser audio. Complete the setup check shared by email before the workshop. No paid subscriptions are needed.
03
What you take away
A runnable booking agent, backend and audio checks you can rerun after every change, and a troubleshooting record with a short runbook.
Two days. One case, start to finish.
Each day pairs guided sessions with hands-on labs. Timings are in the schedule below.
Day 1
Build a voice agent that can book
Follow a browser conversation from microphone input to a confirmed booking.
You leave with
A voice agent you can call from the browser.
A correct booking after the caller changes their mind.
A session ID that ties the call to the backend record.
Day 2
Debug the call that goes wrong
Investigate a slow response and recover from a booking timeout.
You leave with
Exactly one booking after a timeout, and an honest reply.
A trace comparison of a normal and a slow call.
A short runbook for the failures you worked through.
The schedule, day by day.
New dates to be confirmed
Build the booking agent
Opening demonstration. A live browser call with the booking agent. What should happen when the caller changes the requested slot?
10:00–11:00
The audio path, end to end
Follow a call from the browser through the LiveKit room and Python worker to speech recognition, the language model and speech synthesis. See how streaming and asynchronous tool calls keep the conversation moving.
Guided session
11:00–11:10
Break
11:10–12:10
Turn-taking and safe bookings
How turn detection, barge-in and pauses shape a call. Walk through availability lookup, slot validation and confirming with the caller before the agent writes a booking.
Guided session
12:10–13:00
Run the starter
Open the local room, start the agent worker and complete one microphone-to-agent turn. Record the session ID.
Hands-on lab V1
13:00–14:00
Lunch
14:00–14:45
When the caller changes their mind
Predict what changes after a spoken correction, and see how to check the final slot in the booking API.
Guided session
14:45–15:30
Wire the tools
Connect the availability and booking tools to the sandbox API. Require a valid slot and explicit confirmation, run one normal booking and record a baseline timing.
Hands-on lab V2
15:30–15:45
Break
15:45–17:00
Test a corrected request
Interrupt the agent mid-reply and change the slot before confirming, then inspect the backend and record what was booked. If time allows, check that two sessions stay isolated.
Hands-on lab V3
Use-cases and examples may be updated. The core content and learning outcomes stay the same.
Times are in IST.
New dates, same hands-on experience.
The workshop is moving to the third or fourth week of january 2027. Registration reopens once the new dates are confirmed; express your interest and we will write to you first. For group bookings or questions, email the program team.