LH129 Ripping Scorm with Mike Alcock

Employees spend almost 2 hours daily searching for information across LMSs, intranets, and document repositories. And large companies have invested millions in courses, LMSs, training libraries, videos, SCORM files, and more. And when the knowledge is needed, it can't be found. It's really difficult. Searching SCORM is close to impossible. And Talvi makes all of that content searchable.

If you're in L&D, Learning Technologies UK 2026 belongs in your diary. 29th and 30th of April this year, it's the UK's leading event for technology-supported learning, co-located with HR Technologies UK and bringing together 15,000 learning and HR professionals. There's a free exhibition with more than 200 exhibitors and 200 seminar sessions alongside a dedicated conference for L&D leaders. That focuses practical insight, peer learning, and examples from organizations including Google, Intel, and Merck. Bookings and registration are open now at learningtechnologies.co.uk.

This podcast is supported by Synthesia. Training videos shouldn't take weeks to make. Synthesia lets you turn text into professional multilingual training videos in minutes. It's fast, scalable, and so simple anyone can use it. You can write an email, you can make a video.

Hey there.

My name is Alex, and I'll be guiding you through this training session.

70% of Fortune 100 companies use Synthesia to modernize onboarding, simplify compliance, and scale global training. Podcast listeners get an exclusive 30-day free trial. Go to synthesia.io/greatminds. Don't miss it. This offer won't last. Welcome to The Learning Hack, a podcast about the people and technologies creating the future of learning. I'm John Helmer.

Now guess what? Learning is cool.

Learning is cool.

Learning is cool!

I'm learning!

Learning is fun! Knowledge is power!

Knowledge.

Education.

Learning. Here's the problem.

Your organization has spent years building a learning library. Courses on compliance, onboarding, product knowledge, systems training. Hundreds of them. All sitting in the LMS. You can enrol in them, click through them, complete them. Trouble is, at the moment, that's pretty much the only way you can get access to that content and the knowledge it contains. If you're a learning professional managing a huge content library, that makes for a hideous ongoing headache. And if you're a learner wanting to get a specific piece of knowledge from that library, which will help you solve the problem you have in front of you today, you're basically stuffed. My guest today thinks he has the solution to this problem in a product called Talvi, which, to put it rather brutally, rips SCORM. Mariana, who is he?

Hack Facts.

Mike Alcock is the founder and managing director of Talvi, an AI-powered corporate knowledge delivery platform. He studied civil engineering at the University of Sheffield before finding his way into learning technology. A journey that took him through an insulation factory in Newcastle and into a career building software products that keep arriving ahead of their time. He co-led GOMO as Managing Director under Learning Technologies Group, and before that founded Atlantic Link, where he built the world's first cloud-based e-learning authoring tool.

In our conversation, Mike takes me through exactly how Talvi does what it does, including why cracking open a SCORM file is a harder problem than it might sound. He's also got some strong views on what happens to the LMS when AI can answer the question faster than any course could. And if you're a vendor with an LMS to sell, it's a bracing listen.

So, Mike, great to have you on The Learning Hack. Welcome to the show.

Thank you, John. Pleasure to be here. Looking forward to the conversation.

Mike, I should tell you that I turn down a lot of requests from learn tech vendors, sometimes 4 a day, to come on this show. I only say yes when I think someone genuinely has something new to say. I think that about you and your company, Talvi. But to be honest, I had a bit of trouble at first getting my head around what Talvi actually does. And I asked some of my old friends in the industry this, you'll probably know a couple of them, in an informal setting, i.e., in the pub, they said, oh, it rips SCORM. And I sort of understand what that means, but only sort of. So I'd be grateful if you could start by telling the listeners in plain language what the problem is that Talvi is solving and what ripping SCORM actually means.

Sure. So the problem we're solving is access to knowledge, especially in the moments of need. So employees spend almost 2 hours daily searching for information across LMSs, intranets, and document repositories. And large companies have invested millions in courses, LMSs, training libraries, videos, SCORM files, and more. And they're fine for sort of formal learning environments, but most of that learning is forgotten pretty quickly after the training course has happened. And when the knowledge is needed, it can't be found. It's really difficult. Searching SCORM is close to impossible. Searching SharePoint is very limited in its capabilities. And Talvi makes all of that content searchable, including SCORM files, but also videos, PDFs, anything that lives on a computer. And it allows people to get instant answers from approved sources in seconds via a simple question. And the SCORM ripping is basically decrypting SCORM files so that they can become searchable and people can get answers from them without having to open them and without having to click next to continue for 30 minutes to find the answer.

Okay, so to make that absolutely clear, a SCORM package, for people who don't deal with this every day, it's a way of kind of packaging up a course, and that might have all sorts of contents in it, sort of audio, video, text, graphics, all the rest of it. And from the outside, you can't really see what's in that SCORM package. And the SCORM package has been the kind of default unit of e-learning, if you want to use that slightly creaky term, for maybe the last 10, 20 years. So your technology allows people to look inside that. So I can see what you're doing is generating a bit of a buzz. It's very timely because people want to bring to bear, you know, the weight of newer technologies like AI on all that information. And there's a bit of a kind of, not a firewall, but a wall there. But I want to push on the practicalities because SCORM packages are not all created equal. Some are essentially slides converted to web pages with lots of readable text. Others are highly visual, heavily interactive. Some are built in Flash or Captivate years ago. You might have been guilty of putting some of those together when you were at Atlantic Link. But How does Talvi handle that diversity? Because, you know, there's a lot of differences within that SCORM thing. Also, there are other companies working in this space. I'm thinking of people like Filtered and their SCORM intelligence product. So how does what Talvi is doing, how different is that from what's already out there?

Sure. I mean, we've got some very advanced capabilities inside Talvi for decrypting SCORM files. A lot of it's our proprietary technology, but we can basically accurately parse and synthesize and extract content from every SCORM file that we've encountered so far. And I've been throwing some of the most horrible old 20-year-old, you know, Captivate Flash SCORM files at our developers. We've cracked all of those, Articulate, Rise, GOMO, all the usual suspects. And you're right, they're all packaged differently. They all behave differently and we have to adapt different techniques. Some of them are very structured and it's easy to get things out because the text is in a separate set of folders from the images, which are in a nice set of folders from the videos and so on. Others, like you say, they're compiled and they're really well encrypted. And we have to use techniques such as visual pattern matching to go and actually look at the slides and extract the text visually from there. But so far to date, we've got 100% SCORM coverage. I'm sure we're going to bump into a random vendor soon where we'll find something that we don't fully understand yet. But based on our experience, we would expect to be able to open any SCORM file and decrypts it fully.

It doesn't sound like it's plug and play. It feels like there's quite a bit of human intervention and expertise involved in that. It's not purely a technology solution.

Yes and no. So for our developers, there's been a lot of human expertise, but for our customers, there's none whatsoever. They upload or we upload on their behalf. Whatever SCORM content they've got, and it automatically decrypts it. Now, yes, our developers have had to spend a lot of time, you know, figuring out the difference between a Captivate, a Rise, and a Storyline, but we've done that and that's built into the Talvi engine.

Oh, okay. Okay. So that's, that human expertise goes into the development rather, you know, try to work out what, to what extent you're a product and to what extent you're consulting, as it were.

It's absolutely a product, product-led pitch. And then to answer the second part of your question, And the difference between, you know, Talvi and some of the other people that are working in this space, that there's quite a few, I'll highlight the top, probably most important ones that we see. Yes, we've got SCORM decryption 100%, but we also have the same percentage capability on just about anything that lives on a computer. We can decrypt video, audio, PDF, and we can decrypt them from any repository. So they could be in SharePoint, Dropbox, Google Drive. So We are synthesizing when we create AI-powered answers from the questions people ask. We're not just interrogating SCORM files. We're synthesizing answers from the entire corporate repository that the customer lets us look at. Next one we do is, um, we have built-in players. So if a cus— if, if somebody asks a question, we give them the answer in seconds. But if they want to verify the answer, they can launch the SCORM file at the exact page where the answer sits. So if we've interrogated a SCORM file to give them an answer, we'll launch the SCORM file at that page. If we interrogated a video to give them the answer, we'll launch the video at the exact second where the question is answered. So the players give people the certainty that the answers they're seeing aren't AI hallucinations, not that our system hallucinates, but they are verified and we present the verified sources.

This is driven by the user, the learner.

Yes. So the user will ask a question, they'll see the answer and they might go, I'm confident in Talvi. I've been using it for a few weeks now. That's the answer I needed back to, back to my work. Maybe the first time they use Talvi, they go, okay, there's a good answer. Sounds about right. Let me click the more details button. I'll read the more detailed answer and let me open the SCORM file or the video and verify the answer. So they've got that option to do it in the flow of work. Next thing we do differently. Yes, we're using AI. Um, but we use, we use RAG technology, vector databases, and intelligent chunking of data. So when a user asks a question in Talvi, Talvi searches everything it's got access to, and it identifies the 100 top potential candidates' files that might have the answer. That might be a combination of SCORM, PDF, and video. On the second pass, we whittle that down to 20. So the 20 top candidates now that are coming up from the advanced search. On the third pass, we identify the top 5 chunks of data that we've got a 99.99% recurring certainty that the answer sits in there. And at that point, we pass those 5 chunks over to the LLM, and the LLM, whether it's ChatGPT or Claude or Copilot, we don't care, generates the answer. The beauty of that approach is that we consume 95% less AI tokens than if you'd uploaded an entire PDF or video or SCORM file to an LLM. So we are incredibly efficient and incredibly cost-effective. So for organizations that are thinking, blimey, if we roll out AI to 10,000 people, just think of the token costs, will be 5% of what they might have expected. And again, for organizations that are looking at their environmental credentials, you know, AI consumes huge data centers and huge amounts of CO2, and we're doing it at 5% of of traditional AI techniques. A couple of others, if it's okay. Um, go ahead. We have got, um, our AI search will automatically identify out-of-date files, old files, uh, which big organizations have probably got thousands of scattered across their LMS and their intranet. But we also have an AI data cleaner, which will identify them before Talvi starts searching. And we've got a human in the loop in that component so that if Talvi can identify, here's two SCORM files, they're very similar, they're 90% the same. Now the admin might need, might not even know that they were both in the LMS, but they can bring them both up, sit them next to each other. Talvi shows them where the differences are and they can manually say, ah, that's the old one. I'm going to delete that one. So you're cleansing your data before the AI starts asking questions. So the AI isn't getting answers from out-of-date files. Last one to tell you about, Talvi Nuggets. Once you've got an answer, let's imagine you've built— your query is, how do I handle payslips? Talvi comes up with an answer, answers your question, and it shows you the 4 SCORM files that it pulled the data from to generate the answer. As an L&D person, you might say, okay, well, we've got 4 bits of information hidden away in the depths of 4 different SCORM files about payslips. But clearly everybody is asking from payslips about payslips. We can see this from the analytics. Let's build a new piece of content about payslips to fill that knowledge gap. Now, the problem you might have is that SCORM file 1 was built 10 years ago in Captivate and you've no longer got your Captivate license. SCORM file 2 was built 5 years ago in Storyline. The person that wrote it's left. You've lost the script, you've lost the video files, you've lost the transcripts and so on. So it could take you 3 or 4 weeks to, you know, just decrypt your own information and create a new piece of content about payslips. Or you could click the Talvi Nuggets button and Talvi will synthesize the payslip information from those 4 files into a brand new piece of learning content. So as well as saving end users hours and hours of search time, we can save the L&D department hours and hours of, you know, curation time to put together new content. In seconds.

That's interesting because all those issues that you're talking about are things that I've heard CLOs, heads of L&D, tearing their hair out about over the years. So it, you know, there's obviously a lot of kind of utility there. So let's say a head of L&D is listening to this right now. They've got a library of maybe 200 SCORM courses, some good, some terrible. Some from 2009, even older. What does the journey look like for them from where they are now to having Talvi working in their organization? What's the integration story? What's it cost them in time and effort? And what's the realistic outcome? I think we're talking a lot nowadays about process re-engineering. To what extent do they have to re-engineer their workflow in order to use Talvi?

So we try and integrate Talvi with their workflows. Talvi is available through a web browser. It's available with voice input so people can access it on the phone. It's also available as a chatbot that can be embedded inside the flow of work. So it could be inside Teams, it could be inside SharePoint, it could be on their website, it could be on their LMS. So there's lots of different ways we can integrate Talvi with their existing workflows so that it's available for people to ask questions in the flow of work. If you're a head of L&D, you've got 200 SCORM files, they're probably sat in an LMS. So from our perspective, ideally we would connect to that LMS via an API. Most half-decent LMSs have got an API. So literally once we've got the connection established and we've got established connectors for lots of LMSs, and if it's an API, it's relatively trivial to build a new one if we haven't encountered it. And literally within 24 hours, we will have ingested, parsed, vectorized, and made searchable 200 SCORM files. To be honest, we'd probably do 2,000 SCORM files in 24 hours. So we've then got a searchable database. Now, for the head of L&D who's got some good and some terrible files from 2009, they may want to spend a week or two going through that process I just described where the human in the loop can look at the, you know, the two files you've got on money laundering side by side and say, well, this is the current one and that 2009 one, let's mark it for deletion because it's clearly out of date. It references, you know, money laundering procedures from 15 years ago and so on. But typically they would be up and running between 4 and 8 weeks from starting work with us. And then the outcome would be Talvi available to their entire employee base. And the savings, like I say, would probably be at least 2 hours a week or more for employees searching with instant access to the content that's locked and buried away on page 58 of a, you know, 10-year-old Captivate file.

Right. So, I mean, it's obviously very useful to people. So you mentioned that some LMSs don't have APIs. Is that the hiccup? Is that going to It's not the end of the world.

It's, it's easiest if they've got an API, of course. We can use webhooks to get into some LMSs as well. Worst case scenario, we have to take a dump out of the LMS, literally extract the SCORM files, ingest them, parse them and search them, and then probably set up a process so that every time a new SCORM file is added to the LMS, it also gets added to Talvi. So it's solvable, but the API makes it really easy because with the API, we get direct connection, we can ingest automatically all of the SCORM content, and then we just set up a listening service through the API. And if there's a new SCORM file added, it goes in, you know, Talvi then grabs it and makes it searchable.

If you're looking to connect with the top minds in learning and talent, there's no better place than an iVentiv event. Since 2009, iVentiv has hosted over 300 events events for global heads of L&D and talent. I've attended multiple iVentiv events myself and found them to be invaluable—less about sitting and listening and more about real, impactful conversations. It's an unmatched opportunity to collaborate with leading chief learning officers on the challenges they're tackling today. Join the conversation! Visit iVentiv.com/learninghack.

That's i I want to dip back into your history briefly because I think it's relevant context. You studied civil engineering at Sheffield University, not the most obvious route into learning technologies. So how did you end up there? How did you end up in learning technology from civil engineering?

Yeah. I did, I did civil engineering at Sheffield. Um, I probably got an engineering kind of mind, but I really found myself towards the end of the third year thinking this probably isn't the career for me. I'm not quite convinced that this is for me. And I literally drifted around from, you know, some graduate jobs trying to find my niche and trying to find where I would fit into the world. And I ended up working in all places of an insulation factory in Newcastle that made loft insulation. And because I was the only graduate in the building, when a computer-controlled production line was installed, they said, right lad, figure out that computer and drive the production line. So I taught myself programming to try and program a computer-controlled production line. And as I started doing it, I thought, well, this is easy. I like this. I can do this. So I predicted for myself a career in programming, but after a year or two of doing programming, I was a little bit bored with it and thought there's more excitement to be had in the world than staring at a computer writing code. So I realized I also preferred designing solutions to problems rather than programming them. So I thought, well, maybe I can get into a job where we're designing solutions. So I ended up working for a company that was a bespoke e-learning content producer. And that was my introduction to the e-learning world, if you like. And I was selling bespoke e-learning, so, and I was helping with the design of solutions, which I quite enjoyed. And when I set up Atlantic Link, uh, you know, with my business partners, we thought we were going to be another, um, content house, another mini Epic or Line or one of those companies that existed at the time. We thought we were going to create courses. And our first customer, God bless them, was Next Directory. And we were working with them for about 6 or 8 weeks trying to create a load of IT training for them. And as we went through this journey, we realized that we was, it was such a horrible manual process. And we sat down in a meeting with NetSaid, look, you don't really need us creating the content for you. What you need is some software that will create the content. You know, that's the answer. And that's going to revolutionize what you're doing and what we're doing. And they said, of course, it's really good. Yeah, you're absolutely right. Can you guys create software? We said, yeah, of course we can. And we walked out of the meeting and said, well, how do you create software? And that was the start of Atlantic Link's journey. And basically we thought, okay, there's an opportunity here to become a software company if we can figure it out. And we got very lucky and found a development team in Kyiv in the Ukraine. It's a long story and I won't go into it here, but we found a development team and suddenly developed something. And I discovered that with my engineering background and combined with a little programming knowledge, it allowed me to speak to developers in a very specific and precise way. And it allowed them to develop exactly what I wanted without confusion. So we suddenly had software coming out that was exactly as I'd had it over a conversation, whether it was on MSN Messenger or via email, and the software would come back absolutely perfect. And that sort of probably leveraged my background as a, as a civil engineer, quite precise instructions that we gave to developers. And that was where we got onto the software journey.

Right, that's really interesting. I mean, a lot in there. Going back to the beginning of your answer, it is interesting that people that I interview come from so many diverse angles into this industry. And yet what you described there as being the person in the company who is looked to as somebody who could help other people make that journey from, you know, into having greater knowledge. Comes up again and again. It's like, well, people just identified me as somebody who could train people and help them with their knowledge. And that was the route in. So I mean, it's never occurred to me as a common thread before, but I mean, that's definitely there. Also, clearly you have a kind of product development mindset. You look at a problem and I have to say that Having worked in a bespoke e-learning company myself in those years, we kind of knew that, well, you know, here's a time management program for this retail bank. Here's the same thing packaged differently for that retail bank. Yeah. It was very much about turning the handle. And, but somebody with a mindset like yours says, you know, how can we kind of make this process all a bit more efficient? And you've carried on. And carried on in that bent. You went on to build what's credited as the world's first cloud-based authoring tool at Atlantic Link, then GOMO, then Instilled, repeatedly being ahead of the market. Associate by your own account, I think that's a bit harsh. That's Claude that put that in. But definitely, I mean, in a sense, if I'm going to be kind made about it. It's a conservative market, learn tech, and it's not hard to be ahead of. Agreed. But which is not to decry your own achievements, which are considerable. But what's the thread that runs through all of these for you? I mean, I'm pulling things out, but I'm interested in your— curious about how you think about it. Is there anything in the engineering mindset specifically that shows up in how you approach building learning products, do you think?

Yes, possibly. I mean, I agree with some of the points you made in the question there. I really think the L&D market has been quite poorly served by software vendors. And when we started with the idea around Atlantic Link, the authoring tools that were around in 2000, I think it was Director, Macromedia Director, Authorware, Lectora, I thought they were horrible.. And, you know, we looked at using those to develop content for Next and we thought, well, these are garbage. Let's surely somebody can do something better. I thought they were clunky, unfriendly, fiendishly difficult to use. And the output was less than optimal to be kind to them. So we thought someone can do it better. We decided to have it, you know, let's give it a go. And I don't know if it was my engineering background as such, but I did see the internet coming and I was talking to my developer in Kiev. And every other tool in the market at that time was a desktop installed tool like Microsoft Word. You know, you installed it on your PC and you worked off it on your own. And the internet was just arriving. And we said, well, the internet's probably going to be big. It looks like it's something that's going to explode here. Can we stick this product into the internet? So, um, we bounced that one around for a while and Vadim, my developer, said, well, yes, we can, but web browsers are not really capable of of running an authoring tool inside them. You know, they literally had view capabilities only at the time. But we, we explored it further and came up with a solution based on the .NET framework, which allowed us to install a client on a PC and connect to a web server. And all of a sudden, we had a collaborative authoring tool, the first one in the world. So a bit of good design, but a bit of, this sounds like a good idea, and a little bit of luck, I guess. But it was interesting that as soon as we did that, we had the cloud-based space to ourselves for the 8 years that we owned Atlantic Link until we sold it. And I just thought the other vendors were either too lazy or complacent to try and put their products into the cloud. You know, Articulate didn't hit the cloud until they did Rise in 2015, 12 years later. Adobe are still selling Captivate today. It's a desktop tool from the 1990s. So, um, yeah, I think we, we got ahead of the curve by doing things that nobody else was doing. Uh, and it took a long time to catch up and to be honest, you know, Lexor would introduce version 9 with a load more menus, um, and a few new features and we'd go to the pub and think that's another year without any competition because we were still the only ones in the cloud. Um, so yeah, we had a great run with it, but on the, on the engineering side of things, uh, again, ability to talk to the developers and get them to understand the vision of what I wanted and, you know, have no sort of questions or queries, being very precise. It allowed us to make things right, like you do when you're an engineer, but right first time. And that really came through. And so the software that came back, I could almost guarantee it was exactly what I was expecting, what I wanted. And I also felt I had a reasonably good view of of what the market wanted and how we could be smarter than competitors. So hopefully that's where the engineering bit came through.

I'm hearing here not only the engineering mindset, I'm hearing the entrepreneurial mindset. And when you're talking there, it so reminds me of the feeling that you would get from a lot of of, well, a group of people in the industry who are forward-looking, like yourselves, where this kind of, we're waiting for the bandwidth to increase. We know this is coming. We know the bandwidth will increase. Will it be this year? Could it be next year? And then we can do it. And on the other hand, yeah, we are waiting for the market to catch up with this. We're waiting for the buyers to realize the advantage here and to start adopting. To a degree, that process continues, but the pace of acceleration in AI is somewhat different. But rather than kind of falling into that, let's talk about AI for an hour rabbit hole, I want to talk about the destiny of kind of learning systems, as it were, based on what you've said so far. Because if what Tavi does becomes widespread, If AI can surface the knowledge inside any course library on demand at the point of need, what happens to the LMS? Does it become redundant or does it find a new role? And more broadly, what happens to the instructional designer who spent years crafting carefully sequenced SCORM courses? Are they still adding value in a world where the AI just serves up what the learner needs when they ask?

Yeah, I think the LMS in its current guise is on borrowed time because of the rise of AI. I suspect there will always be a need, but again, AI may eventually replace this. But I think there'll always be a need for, especially in compliance, to prove to, you know, HSBC or the major banks, they've got to prove that 100,000 people have done their money laundering course once a year. So you've got to have a source of record for that. Now that might become an AI-powered proof of competence or proof of actually just sitting through the course and so on. And I think the LMS has also got some places in things like learning paths, but again, I think AI is going to replace that. So, yeah, if I was an LMS vendor, I know they've all stuck a powered by AI badge on the front of their logos and websites, but you know, I'm trying to think of the analogy and it escapes me at the moment, but yeah, a lot of these are non-AI native products and I think they're going to struggle and I think they will be replaced when people realize that the answers can be served up in seconds without having to force people to click next to continue for an hour and a half to get to the content they need. So I think compliance potentially, maybe there's a little bit around structured learning paths, but again, I just think conversational AI interface will be able to provide that to people very, you know, with guidelines and rules. I think AI will take that over. So I suspect it might just be compliance in the future and the traditional LMS might disappear and most people will go through a chat-based interface with an AI coach or helper or assistant working with them. For the learning designer, I think they're likely to become more design and curation, curation rather than creation, if you know what I mean. I'm seeing authoring tools now where you can give it a prompt or two and it will virtually write the entire course and it'll source the imagery and it'll write the paragraphs of text and it'll set up the quiz and all of that sort of stuff. And that's the stuff we were trying to make easier to do with Atlantic Link 20-odd years ago. It can all be done at the touch of a button. In fact, one of the things I was offered before I settled on building out Talvi was to build an AI-powered authoring tool, But I just thought, firstly, it's a relatively trivial task, but also, yes, it can be done and yes, it's relatively easy, but again, you're commoditizing more click next to continue content. And is that what the world needs? So I think the learning designer, the instructional designer is going to be more of a designer of the overall experience and a curator of content and ideas and approaches. Rather than spending their days in Storyline creating another 50 slides to explain money laundering or whatever topic they may need to want to get over.

A learning experience designer, as they've been calling themselves quite a while. These ideas do hang around. Yeah, of course, I should say the death of the LMS has been heralded since about—

It hasn't. You know, I did produce an article on the death of SCORM, and it really It's 25 years old and at some point it's going to go, but yes, it'll probably be here long after I'm gone. Yeah.

Well, particularly the history of media, everybody thought that radio would disappear with television and so on. It doesn't. We still have everything. Interestingly, in thinking about this, I think the CEO of Atlassian talks about— you mentioned a system of record,— a contrast between systems of record and, I think you called it, systems of knowledge. And maybe that's a useful way of thinking about, as the LMS persists for certain uses, you know, what else happens in reshaping that market. Anyway, something else you touched on was the idea of learning in the flow of work, getting knowledge to people at the moment of need. It's been one of the most discussed concepts in L&D for the best part of a decade. Claude thinks it was Josh Berson's idea, but he did bring it to fame, as it were. And I should say that our next guest on The Learning Hack is Bob Mosher, of course, friend of the podcast. But what we hear from people in the field has often been, and now I'm going to be devil's advocate and put the argument against learning in the flow of work here. Difficult for me because, as I say, I'm a good friend of Bob's. Sounds great, but it doesn't really land. People don't stop what they're doing to learn. They work around knowledge gaps. They ask a colleague, they Google it. If workflow learning is part of what you're promising with Talvi, and actually looking back over those years since Berson announced learning in the flow of work as the new paradigm, it hasn't caught on massively. It's certainly not mainstream. What makes Talvi different from the chatbots and knowledge bases and all the just-in-time learning tools that came before it? And if the friction is still fundamentally cultural and organizational, which it seems, can any amount of clever AI really solve that? Yeah, that's a really interesting question.

I think the big difference is that the AI-powered search that we are able to implement inside Talvi is so fundamentally different from the kind of search that the old chatbots, knowledge bases, and previous attempts at just-in-time learning were able to do. Yeah. So we use a whole new raft of technologies, retrieval augmented generation, semantic and keyword search, vector databases to deliver highly accurate answers in seconds from approved data sources. And we're doing this deep dive into SCORM and video Most of those things were not possible before the advent of AI. And so what we do when people do a search now in Talvi is we get users to an answer, whereas the old learning in the flow of work might have got people to a list of documents or courses or videos. But you haven't really answered the question, you've just provided a list of sources. And I think it's a bit like the comparison between ChatGPT and Google today, just pure Google rather than the AI bit. If you type a question into ChatGPT today, you know, tell me about the latest Ferrari Dino, for example, it will come back with a big answer about the Ferrari Dino, and it might bring up some photographs and so on. It answers the question. It tells you all about the Ferrari Dino. And if you type that question into the old version of Google, it might come up with the Ferrari website, some fans website, a list of 20 websites. It doesn't really tell you about the Ferrari Dino, and you might spend your time clicking on 20 websites before you to the bit you're interested in. So AI search and what we're doing in Talvi is so fundamentally different. I believe that we're now at the true point of being able to do learning and knowledge in the flow of work. And it's night and day compared to those chatbots and knowledge bases from 5 or 10 years ago. So I think culturally and organizationally, organizations are looking at what we're doing now and say, oh, hang on, on. This appears to have solved that problem. Yeah, we've been hearing it for 5 to 10 years, but this is actually a solution that we can, you know, roll out to our users and be confident they're going to get the answers they want in seconds.

I've got to ask you to explain a couple of the terms you use there. A lot of people will be very familiar with RAG and vector databases. Other people won't, and I'm conscious of the fact they may have been sitting through this interview, if they're still with us, not having understood by what you meant. Can you give us a quick explainer on both of those?

Yeah, RAG, Retrieval Augmented Generation, it's an advanced AI search technique, which does more than just standard search. And it does things like relevance and scoring, and it will look at documents and say, okay, you're a 10-year-old PDF and no one's looked at you for 5 years, Therefore, I rank you lower than the more modern PDF that's, that's been accessed recently and updated in the last 2 weeks. So it's, it's more intelligent search. So the old search would have found both of those documents and identified them with the same score. RAG allows you to push them down there. Within that, we have semantic search as well as keyword search. So when we're searching through documents, we We've searched on the meaning of the phrase. You know, someone types in money laundering. We don't just search on and produce results with the word money or laundering or money and laundering together. Semantic search says, oh, money laundering, that's a compliance measure. So if I find some interesting content on compliance, I'll include that as part of my answer bank when I start to generate the answers and pass them to AI. Vector databases are just an incredibly efficient way to make content searchable at the speed of light. And basically we turn all the text and everything else that we consume into TalVee into numbers. It's pure numbers inside a vector database. Just allows us to search incredibly quickly and get to the answers. And then we recompile it back into text or whatever form it might be in order for people to be able to read it. It. But when you vectorize stuff, it becomes incredibly fast to search.

Good. Thank you very much for that. That's very clear explanations. I just, I'm kind of curious about how we talked earlier about the speed of change in development, how it's a little slow in learn tech. And we were waiting for bandwidth and so on to increase and so on. And there's cultural issues, there's technological issues. How do you feel about the current very rapid pace of development in AI as a field, which is different because it doesn't move in a straight line, it's exponential? How do you keep up with that? Do you take it as a challenge, an opportunity to get ahead of the market? Excited? How do you react to that as an entrepreneur as well as offer as an engineer?

Yeah, I think it's an absolutely huge opportunity to do things from a software perspective. We can do things at an incredible rate compared to what we did 20 years ago. The first products I built, I had a developer in Kyiv and there was no AI and he was manually coding and writing every single line of code. And a lot of that code writing is now done by AI. We still have humans orchestrating it and I think we always will, because we've got the oversight and the experience to guide the AI. But, you know, the drudgery, if you like, of software development is really disappearing. Where it will go in the future, who knows? But, you know, we've got our finger on the pulse and we will, you know, either follow or lead. But I believe we're leading with some of the stuff we're doing in the L&D sector at the moment. I'd like to see us as a leader in the future as well. But just in my my, you know, from my personal perspective, to give an example, you know, for me in my day-to-day work, um, I needed to get the Talvi website together. And I spend my— I do that entirely in AI. I use a product called Replit, which I really love, and it builds my website for me and it designs my graphics and logos and pages and layouts and everything. Um, and I have a conversation with it, and that's just for me personally and my my day-to-day job, not my developers' jobs. And the amount of time and probably money that saved me is off the chart. You know, and 2 years ago, I would have been paying a web design agency £20,000 for a website like mine. And then I'd be paying them a retainer of £1,000 a month to do my updates. And I'd send them the updates and they'd make a mistake and I'd review it 2 weeks later and say, no, no, I wanted that logo over there. And, you know, I would have iterated that. I have a quick chat with Replit, it. Yep, that's good. Bang. And it's published. So it's a huge opportunity. And I think, you know, for anyone that's worried about AI, it really isn't that scary and try and embrace it because you'll be absolutely astonished at what you can achieve. You know, I'm now inverted commas, a web designer, which is something I never would have expected to be 2 years ago, as well as being a CEO of an AI software company, which I sort of expected. That's it.

Another string to your bow. And do you live in kind of constant awareness of what Ethan Mollick talks about when he says, you know, any solution you design for now is going to be out of date in 6 months' time?

There is that, of course. But again, you know, we've got a— our team are constantly forward-looking. And, you know, I do follow Berson and all those guys just from the perspective of the L&D and what's getting ground attraction in the L&D space. But my dev team are AI native and have got their finger on the pulse. And so we're already looking at what's coming next. So don't ask me what the acronym stands for, but MCP, that's already built into our product. I think I know.

Model Context Protocol. Thank you.

Yes, Model Context Protocol. I'll take that. Sorry. Yeah. So we're already building out those solutions for the next generation, which is going to be agentic AI. We're a multi-agentic framework inside Talvi, and we can now leverage these MCP protocols to do things that, again, we're seeing the more advanced players in the market starting to move towards. And we're hopefully keeping our finger on the pulse and again, trying to stay ahead of the game. I think it's probably a little bit more difficult to stay ahead of the game now than it was in 2002, because no one was really brave enough to try and stick stuff in the cloud and waited for 10 years to do it. Whereas I think now AI makes it a lot easier. So, We've probably got to run a little bit harder than we did back in the day.

I think that, you know, there's a consistent thread of being very farsighted and also quite brave in your story. And I've desperately tried to tease out some negatives and some downside, but you're obviously a very cheery fella. Last question, an occasional ask on The Learning Hack. I never quite know, people I don't know that well, I never quite know how they're going to react to this one. Sometimes they just say, no, I don't do that. But if you were forced onto a karaoke stage tonight, what song is your go-to favorite?

Um, I haven't quite got Liam's voice, so I'd love to do some Oasis because I'm a Manc, and, um, Oasis or the Stone Roses, but I don't think my voice is up to it. So I blag my way through Funky Cold Medina by Tone Loc because you can more or less talk through the rap on that one, and I know it back out. So, uh, Funky Cold Medina by Tone Moke.

Yeah, I think Stone Roses, you know, without throwing any shade, uh, less of a high bar in terms of singing.

Yes, absolutely. Yeah, I think Liam's voice stands up to scrutiny a little bit better than Ian's.

You've got to be able to sing those two notes he uses very clearly. Sorry, I'll shut up about music now. Thank you very much, Mike. Um, it's been really interesting. I've learned something from that and And I'm sure the listeners have too. Thanks a lot for taking the time.

Thank you, John. It's been a pleasure. Speak to you soon. Cheers.

That's all on The Learning Hack Podcast for this time. Many thanks to our guests and to our sponsors. The Learning Hack is completely independent and transparently funded by sponsorship. It is produced by John Helmer and hosted by him and also me, Mariana, a synthesia AI avatar. If you want to help others find us, please like, follow, rate, review, and subscribe wherever you get your podcasts or on YouTube.

We'll be back in your feed in 2 weeks' time.

Meanwhile, stay curious, learning people.

Wow! Now I finally get it! The Learning Hack Podcast is supported by Learning News, the learning sector's newswire. Rob and his team are good friends of the podcast and we really value the help and advice we've had from them, and they do a great job. For the very latest news from around the learning sector, for interviews with learning leaders, the latest from learning sector vendors, and features on workplace learning, go to learningnews.com.