[00:00:00] Nadiyah Rajabally: Hi everyone and welcome to another interview for our AI career, and I'm Nadiyah, head of marketing at hundo, and I've got the wonderful Scott, who's our technical co-founder, and we are gonna be talking about navigating careers with ai. Insights from our very own Career Copilot. So Scott, do you want to introduce yourself?

[00:00:28] Scott Byrne-Fraser: Hello everybody. And thank you for the introduction. Yeah, it's fantastic to be here today. Yeah. I'm Scott Byrne-Fraser. Uh, I'm the technical co founder here at hundo. Been with hundo now for several years, helping to build out the, uh, build out the tech stack and create the hundo. xyz product as it currently is.

[00:00:46] Nadiyah Rajabally: So Scott, let's dive straight into the question. One thing is. That people might not know is what Career Copilot is. Could you give an explainer what it is?

[00:00:55] Scott Byrne-Fraser: Absolutely. So Career Copilot is a AI driven careers guidance tool. So it's an AI chat bot that we've trained on thousands of different organisations in terms of like what they hire for the skills that they hire for, um, their location.

We trained it on how to give. Good careers advice with guidance in terms of how to structure the questions that you ask for people. And really it's aimed to be an always on tool that students or anybody can use to ask questions and aid in their research when it comes to finding a new role, you know, be that their first role out of college or out of school, be it helping them how to prepare for an interview, how to research the companies that they may want to work for, how to write a CV.

It's really aimed at all those different aspects. Um, so it can support you in your sort of day to day career journey.

[00:01:53] Nadiyah Rajabally: Interesting. Yeah. I remember when we, First got Career Copilot, it was so cool, like doing a little test and stuff, and now having it working and having actual students actually using it and getting career guidance.

So in what ways does AI impact traditional career guidance and how can AI based tools such as Career Copilot help individuals find careers that match their skills, interests, and backgrounds?

[00:02:15] Scott Byrne-Fraser: No, it's a really good question. And I think when we first started looking at the concept of Career Copilot, it was really a question as in terms of can, can AI support here and how can AI support?

And we looked a lot of what the types of conversations that careers guidance counselors were actually having, or teachers or lecturers, or even parents with their children and what types of conversations they were having and how we could help support that. Cause I think that a tool like this or any AI tool.

Is really, it's just that it's a tool, you know, so it helps you in that journey. You know, if you are in the space of giving careers guidance, you know, it helps amplify what you do. You know, it helps give your students an always on tool that they can use when they're not speaking to you, but you can still have that one on one relationship with them, you still know the nuance of who they are.

Um, I think that was something also critical when we were building it was that. All the off the shelf tools that we looked at, uh, all the off the shelf, AI tools could give a level of careers guidance, but they didn't know anything about the people that they were chatting with. They had to ask a lot of questions.

So the key thing we did do is we, we use data that's available on the hundo profile that students have chose to share, and we use that to help guide the conversation. So if the AI. But understands what your interests are, what skills you currently have, where you've worked in the past, where you want to work in the future.

It can obviously give you much more specific, um, types of advice. Um, and specifically the skills piece, you know, that's really important is, you know, as part of the hundo profile, we add your skills on that. We add your achievements on there, either achievements that you've. Got before you join hundo or as part of using hundo, they appear on that.

And that really forms the basis of informs the co pilot about what you can potentially do. So, you know, I could say to you, Nadiyah, you know, what are you good at? And you would tell me what you were potentially good at, um, that the record can then. Um, go into the AI and then that, that can really help inform that.

Um, and then the training we did on the employer side was really to extract for what type of skills do employers typically go for. So if you are in, um, if you are in like the green space or you're hiring a particular type of role, what skills are you looking for? And then really we just match them. You know, so, but actually we've got an individual over here who's good at these things.

We've got roles over here that are hiring for these things. Uh, here's how we can potentially match you and then give you guidance as to how you can potentially get there. Um, so it's really just doing a lot of research for you. You know, a lot of that information is available out there on the internet.

It's just really hard to get. You know, if you want to go and research thousands of companies and find out who is actually hiring in your space, you kind of waiting to see if they advertise for a role, um, which you may see, you may not see, um, well, at least with a tool like this, it allows you to kind of speed up that process and get way more information quicker.

[00:05:01] Nadiyah Rajabally: No, definitely. And one important thing that you said, it's true. There's so much information out there. You just get overwhelmed. For me, when I research stuff, I'm like, this is like a lot of information. It's hard to take it in. And it's like, where's the starting point? Where's the end point? It's just, it's endless.

And then having a Career Copilot, having something that's personalized to you and your needs and what you want is so much easier. Cause I know even when we were using it, even when we were asking the students out, um, I think it's South Staff College, and they were saying how, When they were using the tools, they were like, they could instantly find a role, and an employer, and learn more about that, and see whether they want to do it or not.

And I thought like, that's the beauty of Career Copilot, which you obviously don't get with other AI. So the way we've trained it, and you guys have trained it, I think is really good.

[00:05:46] Scott Byrne-Fraser: Yeah, yeah, I agree. I agree. And it was, it was really interesting when we were doing those early, early tests and seeing how people use it.

Cause obviously we had certain roles, we had certain skills in mind, but we can't test everything. But every time we've got a new student using it, who came from a totally different perspective, you know, they had a different background, they had different skill sets and they could still get value from it.

I think that was really, really powerful from our perspective that told us that there was no big potential in using these types of tools. And I think. I think that's really the strength of a lot of AI tools is they can take huge amounts of information and then they can distill it down for you and simplify it and make it easier to discover, you know, interesting insights from that information.

Um, you know, it'd be very difficult for any individual to know about every given career you could potentially do about every different industry you could potentially work in. I just don't think it's that feasible, particularly as things are evolving and changing all the time. So to have a data set you can call on that gives you advice, that gives you Information at your fingertips is very powerful.

You know, if you are a teacher, you are a parent, you are in careers guidance, you know, using the tool yourself to be able to find out and look into research of industries of particular companies of roles, you know, it can help you their own education for them, the conversations you're having with students as well, so it can work in.

So, so many different ways. Um, I think the really interesting thing is because it's such a big data set. It's very rare that you find a type of role or a type of skill that someone's interested in, that it can't give you a steer into the right direction. And then when it does, we do find gaps. It's really easy for us to plug it.

Cause we understand where those gaps are. It's like, okay, it doesn't know much about that industry. Let's find out about that industry, add that info, industry information to the knowledge base, and then we can improve it.

[00:07:25] Nadiyah Rajabally: Yeah, no, definitely. And also one thing is it's like 24 7, so you can just ask it wherever you want, whenever you are, and that's the best part of it.

[00:07:34] Scott Byrne-Fraser: That's it. You can be having your breakfast in the morning, question comes up in your mind, you pull out your phone and then you can, you can ask that question. Um, if you were doing that with your teachers, you'd probably drive them crazy.

[00:07:43] Nadiyah Rajabally: Yeah. No, it's true. So what role should, um, educators play in integrating AI career guidance tools such as Career Copilot into their curriculum and help prepare their students for the future work?

And what benefits do these tools offer? In an educational setting as well. I think

[00:08:01] Scott Byrne-Fraser: starting from the last question, I think the research component to it helps with the education side. So being able to guide students into how to use any AI tool, not just Career Copilot, but how to use any AI tool to be able to supplement the research that they're doing is incredibly important.

Um, It's, it's building on teaching students how to be able to research on the internet, so how to research in a library is about another tool that you can now use to do more research on and also help you with your supporting you in, in your work. I think it's really important that, you know, teachers. And any educators are working with their students to tell them where the limitations are of any given tool about where they need to watch out.

You know, if they start doing all their homework with it, for instance, you know, they're not necessarily then learning the information and they're learning the right tool. So it's, it's really important for teachers to understand. to work with you to help them understand how to use those tools. But then in terms of careers guidance, it's really the same.

It's about, you know, use these tools as a way of giving you a steer in the right direction. You know, it can help you write that CV. It can help you write that cover letter, answering questions and interviews. But, you know, it should never replace your own critical thinking. It should never replace your own imagination so you can input into it.

You can get feedback from it and it's that two way conversation. But I think teachers can be really pivotal in helping students understand that what to do with it next. You know, don't just trust everything that it says, no matter how good the model is, Career Copilot is very accurate. Most of the time, but you still have to be critical of it.

You still have to say, well, is that really right for me? You know, one of my own personal criticism of our own system is it can be overly positive in terms of a role that you're going for. And if, if you genuinely stand no chance of getting a role, it obviously can't say that because it doesn't understand that you won't get that as is, if you are being critical about your research, you can reflect in on yourself and think about where the challenges may be in order to move into a space.

So I think it's really important for teachers to, Uh, and educators to be involved in that way. I'd also say moving forward, there's a, you know, big opportunity for anybody in the guidance space, in the education space to actually be more involved in training models themselves. You know, bringing a lot of the expertise that they have and bringing that into the training models.

You know, if you are a guidance counselor, you are working with students in a particular way. You've got a ton of expertise and you've got a ton of expertise in terms of how you approach it, that a model by itself with the training we have won't necessarily have. So I think you could start to look at how you can then supplement it.

Creating your own models that could then be used to amplify what you as an educator do, or partner with like career copilot to import into them, give feedback to be able to improve them in the right kind of direction. So I think there's many different ways that you can, uh, teachers and educators as a whole can be involved.

But the key thing is just to be, you know, leading the way, demonstrating how to use the tools, demonstrating how to use the best input and into them, feeding back to the likes of us when they see opportunities. And really just guiding their students through, you know, the best way of using what is a very emerging new technology.

Yeah,

[00:11:00] Nadiyah Rajabally: no, definitely. And what about from a parent's perspective? So obviously. They can use Career Copilot at home, anyone that's unsubscribed to hundo can use hundo anywhere. So from a parent's perspective, how can they support their child with AI tools like Career Copilot and helping them get ready for the future?

[00:11:18] Scott Byrne-Fraser: I think it's very similar. So as a parent, you can use the tool to do the research for your child. Your kids effectively, you know, so if your, if your children are interested in particular subjects, you can be using Career Copilot to help guide them in terms of just your own education. You know, I'm, I'm a father, I have a son.

He's not nowhere near ready for going out to get him work yet, but when he is interested in topics, you know, using these types of tools to start to explore the types of, you know, the types of industries he could potentially work in the future, the types of jobs that you could have, children have a lot of questions naturally.

You know, teenagers even more so, you know, they have a lot of questions they want to answer and, you know, you as a parent, try your best to answer them as best you can. You've got your own experience to call on. You've got your, your network to call on. You can do some Googling yourself if you don't know the answer.

Um, so it's another tool in your arsenal to be able to look at and say, well, actually I can help you when you have those questions about your career. Giving you a steer about the types of subjects that you should be doing. Maybe the types of industry you might want to look at. Um, also as a parent, you're inspiring you into a different direction.

You could possibly support, you know, your children in, you know, their interest in a certain area. Um, but there might be opportunities that you would never even thought of because they didn't exist 10 years ago, or they didn't exist five years ago. Ever, ever changing industry. So I think as, as a parent, you're using the tools themselves just to, you know, Research for your children, or even for yourself to understand how to use them, uh, would be the best starting point.

You know, find out what your children are into and use that as a guiding, as a guiding point. You know, I remember, I remember being a teenager asking my parents about, you know, what I would do after my GCSEs, what GCSEs should I do? And obviously they tried their best to answer the questions. I didn't really know what I wanted to do.

I knew, I want to do something with technology, but that was about it. Um, and just having, being able to direct them towards a tool that says, well, say these are the things you're interested in. Here's a five, five or six different opportunities. It thinks that you might be worth exploring. Um, and it can just ignite that, um, inspiration really that can, uh, that can motivate people to then take that next learning step.

[00:13:21] Nadiyah Rajabally: No, definitely. And that's something interesting. Like I said, yeah, if we're like not parents. Aren't aware of like, what's out there, especially any new emerging roles, any new industries. And I feel like Career Copilot really does help that. And any questions, like you said, children, all they have loads of questions.

And I feel like it's an easy way for parents to use these tools or any AI tools and just get the answers so quickly and learn so fast. Why instead of having to research on Google, read all these articles and all these, go to all these websites and get such an easy way and fast way just to learn so quickly and help.

They're young person and give them the right information, at least guide them to the right way. And I feel like, yeah, that's important. So can you explain how you built Career Copilot and some of the challenges and milestones you encountered along the way, and how obviously hundo addresses these risks of obviously using AI tools and making sure that it's ethical and ethically used.

[00:14:17] Scott Byrne-Fraser: Like all good ideas, it started from, well, it started from a very quick Slack conversation. Um, and. I took the idea forward, really just started experimenting. So back in December, so we, it was a very short turnaround between when we first started experimenting with it until when we actually launched it as a feature.

Um, so back in December, we started experimenting with. Uh, what we could do with AI around careers guidance. And I say back in December, but, uh, for a few months last year, we were trying different models, different experiments, just to see where AI had a place, if at all, in terms of our, our products that, you know, did it meet an actual user need for what we, for, for the young people that we work with?

Um, so I tried a few different variants in terms of. Like a CV builder and a tool to help with cover letters, tools to help with research careers, guidance tools. And it started with a really hacky prototype where we took an open AI, open AI API, I hooked it up, uh, through Zapier to Slack. And literally use Slack to start querying really quickly ideas and seeing what responses I got back again, taking data from my own profile on hundo and formatting that data in various different ways, mixing that in with the prompts that we were sending through to Open Ai.

ai just to see what types of responses I was getting, um, when we started to structure that data in a different way. And I got to a place really quickly where the answers it was giving were actually quite useful. You know, when, when I provide certain information to it, um, in terms of taking some information from my profile, as I mentioned before, things like your skills, things like, um, your interest, your job history.

Um, I started layering in some information about employers. You know, it was only four or five employers to start with just to see what the impact was and then started to expand the data set. And it got to a place relatively quickly where. There was the nugget of a potential product that, you know, there was a, there was a potential feature that we could experiment with.

Um, so then the next step was to work out how we could build something, uh, for hundo, but do it in a way that was effectively a lot of safeguards in place. Cause obviously we're working with students. We're working with young people. Um, so we looked around for various different tools. So we use a middle level, uh, middleware.

Um, which we built the, um, the co pilot onto and the middle where allows us to insert our own training model, our own knowledge base, um, around specific subjects. So then we experimented them with various different datasets to see what answers we got. We ended up looking at a, uh, several different datasets, one which was around employers.

So it was around employers, which then broke down into information about the employer, information about the types of roles that employers hire for in a given industry. The skills that are particularly hired for, for those roles in that industry, and then general information about an industry as a. In itself.

Um, and very quickly we got to the point where we had, I think it was 175, 000 different data points between skills and industries and roles, um, that we pulled from various different public sources that we could pull them, we could scrape from the internet, um, but it was all publicly available again, you could in theory, Google it, but because it was, um, it's incredibly difficult to do that, you know, at any kind of scale.

Um, the pieces that we then layered in were, you know, the specific skills mapping, and we actually used separate AI tooling. Um, to create the pipeline, to be able to match the skills to the employers, to the roles in our data set, which we then added to our knowledge base and then that knowledge base, the employee knowledge base and form the basis of it, we then we're looking at the opposite side, which was looking at the hundo profile data, i.

e. the student data they choose to share and how we best structure that to then query and prompt that, that knowledge base. Again, a few different experimentations in terms of like the right level of information, we actually found if we shared everything, it was actually less useful than if we just shared very key specific things like the location you potentially want to work in, the industry you potentially want to work in and your interests gets you an awfully long way.

You then layer in skills and that makes it super specific. And anything more than that actually starts to deflect from it slightly. Um, We also found that well written profiles, well written sort of bios works incredibly well. Um, so again, we try to guide students through creating those. So anyway, we got to the point where we had these two, these different data sets that we're all working in, in the similar model, and then we tested it a lot, an awful lot, just lots and lots of testing to.

You know, make sure that there was no hallucinations in there or whether it was hallucinations. It was easy to pick up on and then rectify them, you know, ensure it was always referring to its knowledge base rather than going outside of its knowledge base to give advice. Um, also given the fact that it is a chat interface, you can ask it anything.

We are working with students, you know, we wanted to make sure that it always brought the advice back into careers guidance area, regardless of what you ask it, even if you go into an area that might be covered by safeguarding. So if you go into an area where you are asking it questions that, you know, maybe you have a mental health issue or maybe, you know, you're, you're struggling to get roll, uh, you're struggling to bring in money, whatever it may be, you know, the tool itself will.

Guide you in the right direction. It will tell you where you can potentially get help either speaking to your educator or, you know, some external resources, but then brings it back to, you know, the careers conversation goes ultimately as a tool. We don't yet believe that AI is the right mechanism to be supporting people in that, that, you know, that type of advice, you know, it's about getting people to real humans to have those kinds of conversations in the safest way possible.

Um, And since then, we've been testing and testing and testing a lot. So there's a test cycle, iteration, iterating in terms of how the model then responds, and you basically build out like this profile of how the AI should talk, and it's, it's now several pages long in terms of what the different rules and everything we have in there in terms of how it, how it responds to different scenarios.

Um, there's a few easter eggs in there as well. If anyone ever finds them, there are a few Easter eggs in there that if you ask it certain questions, it will respond in certain ways, but I won't, I won't disclose those here. They're, um, look at the quite funny ones. Um, but then in a matter of about two or three months, we got to the point where we could, we could launch it.

And we launched the first version of it again, because it's an AI model. Um, we made the decision very quickly that getting it into the hands of actual students quickly was the best way of us being able to get actual feedback about how effective it was. See the conversations that they have, get that feedback, get the, see their responses.

So we could then start to iterate the model. Um, we made a decision very early on that we weren't going to use student data to retrain the model as a whole. So it was, so, you know, students information itself wasn't being used to, uh, in training, train and improve the model directly. Um, because we went to safeguard student data and information instead of abstract it away.

So we look at a broader learning patterns than we use that to train the model. Um, because we don't want to in any way, shape or form pass student data through to an AI at this stage. Um, and that got us to the point where we could launch it. And then, you know, since we launched, we've. We've been through several variations.

We update the knowledge base on a relatively regular basis, um, just to refine the types of answers that it gives. Um, it now supports quite comfortably many different languages. So if you ask it a question in almost any language and ask it to respond in that language, it will do. Um, it's, it reaches now global in terms of the organisations that are on there.

So it now has a global reach. Um, Um, yeah, we're constantly looking at new ways to improve it. And, you know, my, my current thinking is how we start to expand that interface beyond it, just being a text based chat interface and thinking about, you know, what the future of. What the future of having a conversation with a, with a bot actually looks like, you know, be it through audio, be it through voice or be it through actually just taking the data it gives you and abstracting it into a different way into different apps on the products.

Um, cause I think the, as a chatbot is very powerful, but again, that's, that's one way that people communicate. There are many other ways we can use that same data set.

[00:22:23] Nadiyah Rajabally: Yeah, definitely. And I remember, yeah, in the last year when you guys were testing it out in Slack and just seeing it and now that it's a natural product on the platform and young people actually using it, it's just crazy.

Yeah, it kind of blew my

[00:22:35] Scott Byrne-Fraser: mind actually how quickly we got it from, You know, the nugget of an idea in December to a real product to several months later out in people's hands. And, you know, the majority of that time was spent refining it, improving it. Um, and just testing to make sure it could actually work.

And I think with tools like this, it's, it's the best way to learn fast. You know, where, where possible use off the shelf components to build something, replace where needed. Like I said, we made the call quite early on to switch to a, to a knowledge base that we have total control of. Obviously slowed down our speed to get it out, but ultimately made a better product.

So, um, there was, there was lots of considerations, but now it's out in the wild and we can test it at scale. Um, we can learn and improve it a lot faster.

[00:23:20] Nadiyah Rajabally: No, definitely. Yeah. Cause we launched it in, I think, January, I remember as well.

[00:23:23] Scott Byrne-Fraser: Yeah.

[00:23:24] Nadiyah Rajabally: Um, and we did the launch event. It was pretty much 6

[00:23:26] Scott Byrne-Fraser: weeks door to door, 2 months if you allow from the very first test.

[00:23:29] Nadiyah Rajabally: Yeah. So, you touched upon this a little bit, but in your opinion, how do you see AI evolving in the next 5 years and how it will impact careers and the future of work?

[00:23:40] Scott Byrne-Fraser: It's a really good question. I've seen a lot of different conversations about how AI will evolve over the next few years. I think, The bits that I know and can guarantee number one is that how the way that we interface with AI will change and will continue to change AI has been with us with software for many, many years and we'll, we'll continue to be so.

And I think the current methodology of it's either a, a chat interface, or maybe it's a voice interface with, uh, which then translates into chat is obviously a natural step, but I think we'll see more and more applications of it. Driving different types of interface drive, driving different types of UI.

So, you know, more traditional UI where you're inputting information, you're getting a response by being driven by. AI tooling, um, AI guidance, you know, I can see already how the data we get on our platform could streamline the way that you give recommendations to people in terms of like the next virtual work experience you do, or even potentially change the way that content is delivered based on your profile.

You know, so the way that you have, the skills that you have already could streamline Change the makeup of a particular course, or the way that a course is delivered could change based on how you prefer to learn, you know, based entirely off the type of input that you put into it. So I think 100 percent we'll start to see different ways of interacting with AI that aren't just text and aren't just voice.

I think that'll be really interesting. AI will probably continue to improve, you know, I've seen recommendations that we'll get to full AGI in the next five years, 10 years, 20 years. Who knows whether we, when we get to that stage of like human level of intelligence from it. I think the key thing will be is it's the application of it, you know, how we, how we interact with it, how we are using it day to day, there's a really good argument for saying that.

Chat GPT is already good enough for what most people need it for. Um, if you are just using it for helping you with careers guidance, it is already more than powerful enough to do that. Uh, so even if it gets, you know, human level levels of consciousness in the next few years, it won't necessarily make the type of interface better, you know, it's about how you interact with

[00:25:53] Nadiyah Rajabally: it.

No, definitely. I think that's important. I can carry on this conversation. There's so many interesting things that we can dive into Scott, but time is up, obviously everyone, do you want to tell everyone where they can connect to you? Is LinkedIn the best place?

[00:26:08] Scott Byrne-Fraser: Yeah, if anyone wants to connect directly to me, um, find me on LinkedIn.

I'm the only Scott Byrne-Fraser on LinkedIn. So just, just search for me on there. Um, you can also follow me on Twitter. My handle is @Sbyrnefraser. Again, it's just my name. So you can, you can follow me on there. And obviously if you want to check out hundo, just go to hundo.xyz and follow all the links on there.

And you can sign up, get a seven day free trial and just start using this, all the tools we've been talking about.

[00:26:35] Nadiyah Rajabally: No definitely, any parents listening or young person can subscribe to hundo and get 7 days free trial. You can also like, test out a virtual experience and obviously have a play around with Career Copilot.

Um, and obviously educators, we can tailor package just for you so you can book a demo with us. There's You'll see the links down at the bottom on our live stream. And obviously if you can't watch the whole event, don't worry. These videos will be available on demand on hundo's platform and on the website.

It's been lovely speaking to Scott. And yeah, if you guys have any questions for me or Scott, anyone on the team, just send us a message.