Building the future of voice AI at https://teapaw.org/t.co/dvEkP7TQsw
Perth, Australia
Joined September 2019
- Tweets1.9K
- Following217
- Followers348
- Likes2.4K
Pinned Tweet
馃く this should not be possible
A booking experience with payment and... zero buttons!
Language User Interfaces (LUI's) are the next big thing.
@trydaily orchestrating @GroqInc ,@DeepgramAI and @cartesia
In hindsight the Steve Irwin-like voice is a bit aggressive 馃悐
Devilishly simple but genius!
Press 2 to talk to an AI agent.
Everyone thinks integrating voice AI is complex but really it's a few minutes config on a box once your agent is ready.
"Hi, just to let you know we currently have a wait time of X mins. Hold the line or press 2 if you'd like to try our AI agent"
Boom!
The customer's who meet your AI agent have chosen to and will be all the more receptive.
It's like a fishing net for early adopters to power your AI transformation.
It also has the wonderful benefit of giving customers a pathway to getting their problem resolved in periods of high load (and giving you staff a beather).
We have done this for a couple of clients now and for everyone involved it is a much more pleasant experience than pressing a button to move all traffic to the AI agent.
It's 6:02pm. You finally have five minutes to ring the bank.
Closed.
Turns out it's not because they don't want your call.
It's because queue maths makes that extra hour brutally expensive.
Which is why "out of hours" is where every company starts with voice AI.
The bit everyone skips: evals.
No evals, no trust.
Your customers stop trusting the agent, your staff stop trusting the numbers, and the pilot never leaves pilot.
An eval should be calling every hour to verify your system end to end.
Nerdy humour of the day - What do you call a corner lookup issue?
An edge case of course!
I'll see myself out.
The only bug on our rural uber replacement agent was when the person was on a corner.
Ironic on a voice deployment where you expect the voice layer to be the issue.
Google Places found it perfectly but address validation wanted an exact street.
Fix live for our rural golfers tonight.
Our voice AI powers grow!
AI In the Australian Outback - our voice AI now acts as an Uber replacement for a rural community 馃
There is a real transport problem in rural Australia.
They don't quite have the volume to support Uber drivers and the distances can be mean they would be too expensive anyway.
We worked alongside a local business at the heart of a rural town to solve it.
Meddle now owns their phone line.
This was our most complex voice agent deployment and it went live two nights ago.
Our Voice AI handling the messy job of confirming their location, routing to a driver and ensuring confirmation so the customer is not left standing on the side of the road.
The foundation is the voice infrastructure we built at VerticalAI.
The only fully Australian voice infra provider, orchestated by the wonderful pipecat library I am forever grateful for @kwindla and team for building.
We worked with the customer on what they really needed.
We didn't over-engineer.
They had been using WhatsApp groups - we moved them to telegram to make use of their ( @levelsio style wisdom playing in our heads).
We use a custom Telegram app for their drivers and admin team.
Took a day to get that working and then it was evals to confirm the integration would work as expected.
Our agent triages the call and dispatches the job to Telegram.
Driver confirms via button.
Voice agent detects and messages the person.
A driver on roster picks it up. The agent never confirms the ride to the caller until a driver has taken the job.
Nobody left waiting.
Pretty amazing to watch in action.
Also cool to see the platform and forward-deployed working together.
The team nailed it.
You can also see that the pattern is bigger than one customer.
It's the sort of software we could never have imagined a couple years ago and now to see a few-week build act as uber replacement for a rural community is pretty mind-blowing.
Also - some solid learnings from this job - be sure to add callers at varying levels of intoxication to your voice agent UAT 馃槀
If you have a complex voice AI idea in mind reach out to our team.
We're a team of non-nonsense builders who will look at your problem and deliver the right tech for the job.
Our goal is to really drive technology on in Australia, starting from the front lines of picking up the phone.
Ever checked the maths on contact centers? It's brutal.
No, really. It's BRUTAL.
Losing half a 4-person contact center doesn't double wait times.
It makes them infinite.
Not "much longer". Infinite.
..and keeps a watchful eye over your operations.
Once your wait time escalates, our agents jump into action until the queue is back at a level you decide on.
We call it auto scaling for contact centers and it's one of the most effective patterns we've seen for Voice Agents
Here is an alternative to the infamous 996 work schedule that is doing the rounds (9am to 9pm 6 days a week).
I'm 41 and still shipping more than most while raising young kids. So I can testify you can run it without burnout risk over long horizon while juggling family.
Set a single hour every night.
After kids are down if you have them.
After a training session if you don't.
An hour is an easy time block to commit to.
Say 7:45 - 8:45pm.
Go hard for an hour. You'll probably hit a wall on anything mildly complex as your brain will be tired. Don't fight it, just push a bit for a few mins then stop and write down where you failed. Pack it in for the night.
Next morning you'll magically wake up with the solution after a good nights sleep while still feeling refreshed.
Do that 3-4 nights a week and you'll knock down a load of problems while only adding 3-4 hours work onto your week.
3-4 hours that didn't impact your family.
So much you'll even be able to think about having another kid ;-)
