Voice This! Podcast: Episode 7 (Part 1)
Development with Guy Tonye
DISCLAIMER: Don’t listen to this episode hungry.
To our fellow engineers, developers, and curious minds, we dedicate the season finale to you!
Join Millani and Guy as they wrap up the season with the final and crucial component of the software development lifecycle: implementation. From his home country of Cameroon to the exciting tech hub of Toronto, Guy takes us through his background and how he got started in building conversational AI apps. Guy also sheds some light on the challenges he’s faced while coding and deploying apps and shares advice on how to avoid them. Oh and of course, there’s a part two. Guy served us way too many golden nuggets to have only one episode.
Final note: this episode was recorded before Voice Tech Global officially disbanded. Read the official notice here.
Part 1 Highlights
- Guy’s origin story
- Guy’s lifetime project
- Challenges between design and implementation phases
- Responsibilities of developers in Conversational AI product development
- API integration strategy
Part 1 Quotes:
Guy’s origin story
Q: How did you get introduced to Conversational AI?
I was born [in Cameroon] but I grew up in France actually. Fast-forward to now. I moved to Canada, Toronto. I joined a company called Connected, which is a software consulting firm. One of the first projects was to help a new customer that was adopting Amazon, Alexa, and trying to get the product out there. That’s how I started in Conversational AI. I was just dropped in there, trying to figure out, “Okay, what is this thing?”. It’s a speaker you can speak to. It answers, it can play music, and so on. That’s how I started my journey in Conversational AI.
Guy’s lifetime project
[15:15] I have had a lifetime project, which is, I found it unfortunate that, some languages are slowly going extinct. So to explain to people, in Cameroon, …it’s like a whole bunch of tribes. Each of the tribes have their own dialect.
The issue is most people will go and speak either French or English. And when there’s fewer and fewer people off the tribe or more, most of them have left wherever they were the country that they’re like will slowly go extinct. So every new generation, fewer and fewer people will learn about it. And, I found in artificial intelligence and conversation AI a way to maintain those.
Handover challenges between design and implementation phases
Q: We want to know when you’re handing over between design and the actual implementation, what are some of the challenges that you’re facing?
[20:50] 1. Machine learning failure: So the first one is the technology relies on a few components that will fail. The bigger one is machine learning. So whenever you build your application, you assume machine learning would do X, there’s a posibility that it’ll do something else and accounting for how machine learning can fail is sometimes something that is challenging because it’s not thought about early on.
And when there is a handover, one of the thing is the handover it’s like, “nice, you do your thing”. And the reality is like, you’ll come back and be “we need more from you”…but machine learning will behave the way that it’s not anticipate it when you start actually trying things out.
[21:50] 2. Anticipating potential issues with API call: The second thing is everything around dynamic data. Yes. API call, we found one, or we imagined one or we assume one. What happens when the different scenarios for that API?
So let’s continue with our restaurant metaphor for salad. Let’s say, I’m assuming people will need to pick between two toppings. What happens if the toppings they’re looking for isn’t there? What happened if in the entire list of toppings that I offer, they want none of them, all of these kind of, that’s what I call the worst case scenario, which is like “that thing. Oh, it’s an edge case, or it will never happen”. It will happen.
And the questions is, a lot of the time with the handoff, this hasn’t been account for. It’s just “make it happen”. The danger with that is, two things can happen. Best case scenario, the developer come back to you. Or the worst scenario, the developer just make a decision.
Q: What does a handover look like? What’s a typical handover for someone who’s not familiar with this term?
[27:30] I would say the more traditional things is they provid a conversational map, as well as a document that kind of outline some of the specificities “here we want you to use a particular API, et cetera”. They’ll provide that document and engineers will then take that on and use it to actually build the application. That is if the designers on that using some of your tools that help you actually create a map that the developer can almost take and the code is almost there. There is a new style of software that is called code generation software where the designer build and out of that design, it will automaticlly generate code.
Responsibilities of developers in conversational AI product development
Q: I want to know a bit about the software. So, if it provides most of the code, then what would developers be adding to it? What would they be looking for? Or how do they finish that project?
[29:20] 1. API Integration: Let’s say we have our salad conversational AI app, and we want to be able to send on Facebook, a status…Instagram… So if you want to update the status, then that API integration that you will use , that’s where the developer will come in, they will focus on that.
2. Optimizing the language model: Another thing is also optimizing the language model. So based on the system that you use, they may want to review the model that you’ve built to see if we can improve the performance.
3. Set up analytics
4. Deployment: And another thing is also taking care of deployment. So effectively, when you work on with Google and Amazon, you need to publish on the marketplace. So taking care of that or wiring things up so that it works.
API integration strategy
Q: What is your strategy when it comes to integrating APIs? You mentioned that a little bit, but, is it identified in the beginning of the project or what does that look like?
[33:05] Involve engineers from the start: So the strategy that I recommend is, early on, while you’re still trying to figure what is that conversation AI app you’re working on, I always recommend for that exploration of API to happen there. And you are here at the beginning. So all the previous podcast episode, just add an engineer in whatever you were picturing, put an engineer, however, representative engineer, just add an engineer there, so that they really go through all those steps.
And they really help for two things. One is as they explore, they can even find an API that could make your app even better. Just like, Hey, you know, you can actually use this. And while you’re doing your research is like, oh, I could do this or let me prototype and see, my users would want that, right.
About Voice This! Podcast
Conversations with the people who make conversational AI 🎙️Join Millani Jayasingkam and guests as they discuss voice technology, conversation design, AI trends, and the strategy of creating effective conversational experiences. Tune in for first-hand learnings, insights, anecdotes, and sometimes jokes! Say hi and send us your questions at: voicethispod@gmail.com.
Find us wherever you get your podcasts or Listen on Spotify!
