Meet Tom Waszkowycz, Co-Founder and CTO at Inkling, who joined the BotCast #11 to answer our usual questions on building with Voice AI.
Who are you?
Hi, I'm Tom. I'm co-founder and CTO of Inkling, and we're building with the goal of making one-to-one tutoring available to every student. So we're building an AI tutor that works just like a real human does. It talks voice-to-voice, face-to-face. It learns over time where you struggle and how you think, and it challenges you rather than giving out the answers.
How did you end up using voice AI?
I've been a software engineer for the last 10 years, and I've seen the incredible changes AI has brought to the coding world in just the last couple of years. As an engineer, the question is: how do I get my thoughts, the things going on in my head, into the agent's context? Voice seems to be the highest bandwidth way to do that.
I also have a background in tutoring, and that made me think this is a great use case in education. There it's even more important that a teacher and a student understand how each other are thinking, what's going on in each other's head. Voice is the best way to let a student talk through the inner workings of their mind. It's the closest thing we have so far to mind reading.
A moment of joy and pain using voice tech.
In education, the joys and pains are often quite close together. Some challenges make education conversations a lot harder than other voice AI conversations. One example is latency and silence. With voice, you usually want to reduce latency and silence as much as possible. You don't want big gaps after the user stops speaking before your AI responds.
In education that's true on one hand. But on the other hand, silence is very valuable: it gives a student time to think. After you ask a question, you need to give them five, ten seconds to engage their brain. That has often been quite painful to get right.
The joy is when you get that balance right, and you see the outcomes for our users: how much they enjoy the platform, and how voice gives them something they just can't get anywhere else. Students who can't afford their own personal tutor. Students with ADHD or learning difficulties, who don't do well learning in a classroom, from a textbook or from a chatbot. Talking through their thoughts and building up that confidence is really special to see when it works for people.
What lesson would you share with other builders?
Don't treat voice as just an extra layer of interface, parallel to your keyboard and mouse input. Rethink how much you can change your product with it. Your users can now literally talk to your product, and your product can talk back. That means there are lots of things you can do very differently.
Why send users feedback forms by email that nobody responds to, if your product is conversation based? You often know when they're using it and when they may be struggling or frustrated. Why not ask them for feedback right then? What's going on in their head? How could things be better?
Where do you see the industry in 12 months and in 5 years?
Over the next 12 months, I'd expect a continuation of the incremental improvements in the quality and speed of text-to-speech and speech-to-text systems. At the start of the year, we got quite a lot of feedback on how uncanny our AI tutor felt and how much friction there was. It's kind of crazy to see that feedback come down month by month. We're still not perfect, but it's changing remarkably fast. Recently, someone was walking behind one of our users during a session with their Inkling tutor and mistook it for a real person. They asked who the student was talking to, and the student had to explain: it's not a real person, it's AI. So expect that to continue.
The more interesting question is where these incremental improvements add up in five years, and what very different thing they unlock. I think it's a very different way for people to interact with software, and for software to be personalized to users. Not personalized based on what you clicked or how long you stayed on a page. If software can actually talk to users and understand their different ways of thinking, then the website I visit can be very different from the one you visit, and the app that loads on my screen different from the one on yours, because these products genuinely understand how we differ as people and the outcomes we want.
What I'd also like to see in five years is model capability increasing so much that you can do the same with on-device or self-hosted models. Then you get that rich understanding of how a user thinks without the obvious privacy concerns.
INKLING
BotCast powered by https://bespoken.ai 🫶
