1 00:00:00,000 --> 00:00:02,000 Hello and welcome. 2 00:00:02,000 --> 00:00:06,600 So today, we're going to look into a pretty fascinating topic. 3 00:00:06,600 --> 00:00:11,500 Imagine for a moment that you have a stack of documents, hours of videos, 4 00:00:11,500 --> 00:00:14,700 yeah, or even just an interesting web page. 5 00:00:14,700 --> 00:00:19,600 And now imagine being able to turn all that into, say, five minutes tops, 6 00:00:19,600 --> 00:00:22,600 into a real interactive learning experience, 7 00:00:22,600 --> 00:00:29,100 all powered by an artificial intelligence that plays the role of a personal tutor. 8 00:00:29,100 --> 00:00:34,000 That's the promise—very ambitious, by the way—of a technology called Startik. 9 00:00:34,000 --> 00:00:38,000 We have in front of us the documents that describe the brand new version 2.0, 10 00:00:38,000 --> 00:00:41,900 and our mission today is really to see what's under the hood. 11 00:00:41,900 --> 00:00:44,200 Yes, to go a bit beyond the marketing pitch. 12 00:00:44,200 --> 00:00:49,299 Exactly. We're going to break down this promise point by point to see if it holds up. 13 00:00:49,299 --> 00:00:52,799 And it’s a promise that comes at just the right time, it has to be said. 14 00:00:52,799 --> 00:00:55,500 The problem, everyone in training knows it. 15 00:00:55,500 --> 00:01:00,099 On one side, there are trainers who are completely overwhelmed. 16 00:01:00,099 --> 00:01:03,099 They have quality content, but it’s just sleeping in folders. 17 00:01:03,099 --> 00:01:05,199 PDFs, old recordings. 18 00:01:05,199 --> 00:01:10,199 Exactly. And turning that into something engaging takes a crazy amount of time. 19 00:01:10,199 --> 00:01:13,199 And on the other side, we’ve got learners who are bored, basically. 20 00:01:13,199 --> 00:01:16,199 They want interaction, personalization. 21 00:01:16,199 --> 00:01:19,800 Startik is trying to respond precisely to this double challenge. 22 00:01:19,800 --> 00:01:21,300 So, let’s go. 23 00:01:21,300 --> 00:01:25,800 The starting point is this idea of transforming content that already exists. 24 00:01:25,800 --> 00:01:29,800 The sources are very clear about this. We’re not just talking about videos. 25 00:01:29,800 --> 00:01:34,300 You can really start from audio files, documents like PDFs, Word files. 26 00:01:34,300 --> 00:01:35,800 Even simple web pages. 27 00:01:35,800 --> 00:01:40,300 Yes. Versatility seems to be the first pillar. 28 00:01:40,300 --> 00:01:45,800 But where it really gets intriguing is what we transform it into. 29 00:01:45,800 --> 00:01:49,300 We’re not just talking about creating a classic online course. 30 00:01:49,300 --> 00:01:53,300 The documents mention infinite adaptive quizzes. 31 00:01:53,300 --> 00:01:57,300 And even the possibility of generating debates with the AI. 32 00:01:57,300 --> 00:02:02,300 Now that—the idea of debating with an AI based on a document—really catches my attention. 33 00:02:02,300 --> 00:02:04,300 How can that even work? 34 00:02:04,300 --> 00:02:08,300 That’s an excellent question. Because we’re touching the heart of their value proposition. 35 00:02:08,300 --> 00:02:11,300 The idea is to put an end to the content graveyard. 36 00:02:11,300 --> 00:02:13,300 For a trainer, that’s huge. 37 00:02:13,300 --> 00:02:18,300 But for the debate, you shouldn’t imagine it as a verbal duel where the AI wants to win. 38 00:02:18,300 --> 00:02:20,300 The goal is pedagogical. 39 00:02:20,300 --> 00:02:21,300 Right. 40 00:02:21,300 --> 00:02:25,300 The AI will be programmed to take a position, maybe even a contrarian one, 41 00:02:25,300 --> 00:02:29,300 to force the learner to argue by relying on the sources. 42 00:02:29,300 --> 00:02:33,300 Ah, I see. The goal is not to convince the machine. 43 00:02:33,300 --> 00:02:34,300 No, not at all. 44 00:02:34,300 --> 00:02:40,300 The goal is to force the learner to structure their thinking, to find evidence in the course. 45 00:02:40,300 --> 00:02:44,300 It’s a kind of intellectual simulation. 46 00:02:44,300 --> 00:02:48,300 But that raises another question, which is implicit but essential. 47 00:02:48,300 --> 00:02:50,300 The quality of the source? 48 00:02:50,300 --> 00:02:53,300 Exactly. Can the AI turn lead into gold? 49 00:02:53,300 --> 00:02:56,300 The documents don’t say it clearly, but we can infer it. 50 00:02:56,300 --> 00:02:58,300 An AI, even the best in the world, 51 00:02:58,300 --> 00:03:01,300 can’t invent knowledge from a mediocre document. 52 00:03:01,300 --> 00:03:05,300 Right. Garbage in, garbage out, as they say. The magic has its limits. 53 00:03:05,300 --> 00:03:09,300 That’s an important nuance. And for those who don’t have any starting content? 54 00:03:09,300 --> 00:03:13,300 Well no, they’re not excluded. That’s where you see they’ve thought of everything. 55 00:03:13,300 --> 00:03:16,300 There’s a feature to write your course directly in the tool, 56 00:03:16,300 --> 00:03:18,300 but with assistance from Startik’s AI. 57 00:03:18,300 --> 00:03:21,300 Okay. So there are two entry points. 58 00:03:21,300 --> 00:03:24,300 Either you transform existing content, or you co‑create something new. 59 00:03:24,300 --> 00:03:26,300 That’s it. The approach is quite comprehensive. 60 00:03:26,300 --> 00:03:29,300 Right. So you can transform any kind of content. 61 00:03:29,300 --> 00:03:32,300 But all of this rests on the brain of the machine. 62 00:03:32,300 --> 00:03:35,300 What do we know about this AI, the Smarter Startik AI? 63 00:03:35,300 --> 00:03:40,300 Is it just a ChatGPT with another name, or is there something more? 64 00:03:40,300 --> 00:03:44,300 Since the promise is big, we read that it precisely understands learners’ intentions. 65 00:03:44,300 --> 00:03:47,300 That’s the most ambitious statement in the whole document. 66 00:03:47,300 --> 00:03:52,300 And that’s where we have to be careful. Understanding an intention is the holy grail of AI. 67 00:03:52,300 --> 00:03:55,300 Does that mean that if a learner asks a vague question, 68 00:03:55,300 --> 00:04:01,300 like “I didn’t get the thing about photosynthesis,” the AI guesses where the blockage is? 69 00:04:01,300 --> 00:04:06,300 Yes. Or is it just a marketing phrase meaning it analyzes keywords? 70 00:04:06,300 --> 00:04:08,300 The difference is huge. 71 00:04:08,300 --> 00:04:12,300 It’s massive. In fact, there’s a capability described that seems to go in that direction. 72 00:04:12,300 --> 00:04:14,300 Automatic search. 73 00:04:14,300 --> 00:04:19,299 When precision is critical, it’s clearly stated that the AI doesn’t rely solely on its general knowledge. 74 00:04:19,299 --> 00:04:23,299 It goes to look for the answer in the documents provided by the trainer. 75 00:04:23,299 --> 00:04:25,299 And even on the web if needed. 76 00:04:25,299 --> 00:04:26,299 Yes. 77 00:04:26,299 --> 00:04:28,299 And that, technically, is a major clue. 78 00:04:28,299 --> 00:04:31,299 It looks very much like a RAG-type architecture. 79 00:04:31,299 --> 00:04:32,299 RAG. 80 00:04:32,299 --> 00:04:34,299 Meaning “retrieve before you answer”. 81 00:04:34,299 --> 00:04:38,299 It’s not just a simple search. It integrates the document’s information into its reasoning. 82 00:04:38,299 --> 00:04:40,299 And that’s crucial for two reasons. 83 00:04:40,299 --> 00:04:41,299 Right. 84 00:04:41,299 --> 00:04:46,299 First, it avoids hallucinations, those answers that sound plausible but are wrong. 85 00:04:46,299 --> 00:04:48,299 Ah yes, the big classic. 86 00:04:48,299 --> 00:04:51,299 And second, it guarantees that the answer is properly grounded in the trainer’s content, 87 00:04:51,299 --> 00:04:54,299 not in something randomly found on the Internet. 88 00:04:54,299 --> 00:04:58,299 So we really move away from the chatbot model that just gives a factual answer, 89 00:04:58,299 --> 00:05:03,299 and we move into the tutor model that needs to understand a train of thought. 90 00:05:03,299 --> 00:05:04,299 That’s exactly it. 91 00:05:04,299 --> 00:05:07,299 That’s the difference between an AI that gives the right answer 92 00:05:07,299 --> 00:05:10,299 and an AI that helps you understand why it’s the right answer. 93 00:05:10,299 --> 00:05:13,299 And that ties in with another key sentence in the documents. 94 00:05:13,299 --> 00:05:18,299 The AI continuously understands the direction of learning progression. 95 00:05:18,299 --> 00:05:19,299 There you go. 96 00:05:19,299 --> 00:05:22,299 That’s the promise of a differentiated learning experience. 97 00:05:22,299 --> 00:05:25,299 The AI is not static; it’s supposed to adapt. 98 00:05:25,299 --> 00:05:28,299 If a learner stumbles several times over the same concept, 99 00:05:28,299 --> 00:05:31,299 the ideal AI won’t just repeat the definition. 100 00:05:31,299 --> 00:05:34,299 It will offer an analogy, another example, 101 00:05:34,299 --> 00:05:36,299 like a good human tutor would. 102 00:05:36,299 --> 00:05:39,299 We move from a reactive tool to a truly adaptive one. 103 00:05:39,299 --> 00:05:41,299 In theory, yes, that’s the promise. 104 00:05:41,299 --> 00:05:44,299 And what’s quite clever is that it gives the trainer control 105 00:05:44,299 --> 00:05:46,299 over the AI’s personality. 106 00:05:46,299 --> 00:05:49,299 You can configure its response format and attitude. 107 00:05:49,299 --> 00:05:52,299 Yes, you can imagine a very academic tone for a law course 108 00:05:52,299 --> 00:05:54,299 and a much more coaching, encouraging tone 109 00:05:54,299 --> 00:05:56,299 for a public-speaking training. 110 00:05:56,299 --> 00:05:59,299 Which leads us directly to the interface, to the user experience. 111 00:05:59,299 --> 00:06:00,299 Exactly. 112 00:06:00,299 --> 00:06:02,299 You can have the best AI in the world. 113 00:06:02,299 --> 00:06:05,299 If the interface is frustrating, learning doesn’t happen. 114 00:06:05,299 --> 00:06:06,299 Absolutely. 115 00:06:06,299 --> 00:06:09,299 And on this point, the documents insist on elements 116 00:06:09,299 --> 00:06:13,299 that might seem like details, but that say a lot. 117 00:06:13,299 --> 00:06:15,299 You can feel an obsession with smoothness. 118 00:06:15,299 --> 00:06:17,299 They mention fast response speed, 119 00:06:17,299 --> 00:06:20,299 support for mathematical formulas, graphs, 120 00:06:20,299 --> 00:06:22,299 and a very important point... 121 00:06:22,299 --> 00:06:23,299 The display of sources. 122 00:06:23,299 --> 00:06:27,299 Yes, the systematic display of the sources used in the AI’s answers. 123 00:06:27,299 --> 00:06:29,299 Obsession is the right word. 124 00:06:29,299 --> 00:06:32,299 They’re not competing on the number of gimmicky features, 125 00:06:32,299 --> 00:06:35,299 but on eliminating anything that can break the flow. 126 00:06:35,299 --> 00:06:38,299 That’s what we call cognitive friction. 127 00:06:38,299 --> 00:06:42,299 It’s that moment when the interface annoys you so much you lose your train of thought. 128 00:06:42,299 --> 00:06:44,299 We’ve all experienced that. 129 00:06:44,299 --> 00:06:46,299 The hell of poorly designed e‑learning platforms 130 00:06:46,299 --> 00:06:49,299 where you spend more time figuring out the site than learning the course. 131 00:06:49,299 --> 00:06:52,299 The goal is to make the technology invisible. 132 00:06:52,299 --> 00:06:56,299 The simple fact that they’re focusing on that is already an excellent sign. 133 00:06:56,299 --> 00:06:57,299 Absolutely. 134 00:06:57,299 --> 00:06:59,299 And every feature addresses a problem. 135 00:06:59,299 --> 00:07:01,299 Support for formulas, for example, 136 00:07:01,299 --> 00:07:04,299 is a clear signal to scientific and technical fields... 137 00:07:04,299 --> 00:07:06,299 Which are often the poor relations of these platforms. 138 00:07:06,299 --> 00:07:07,299 Often. 139 00:07:07,299 --> 00:07:10,299 Displaying sources is about building trust. 140 00:07:10,299 --> 00:07:12,299 It’s about where the information comes from. 141 00:07:12,299 --> 00:07:15,299 And things like background audio playback on mobile... 142 00:07:15,299 --> 00:07:16,299 Yes, that’s very smart. 143 00:07:16,299 --> 00:07:18,299 It shows they understand modern usage. 144 00:07:18,299 --> 00:07:21,299 You can listen to your course on public transport, while working out. 145 00:07:21,299 --> 00:07:23,299 Learning fits into life. 146 00:07:23,299 --> 00:07:25,299 It no longer puts it on pause. 147 00:07:25,299 --> 00:07:28,299 And there’s this statement that made me stop. 148 00:07:28,299 --> 00:07:30,299 Because it’s so bold. 149 00:07:30,299 --> 00:07:35,299 They guarantee an identical experience at more than 99% on mobile, 150 00:07:35,299 --> 00:07:37,299 tablet, and desktop. 151 00:07:37,299 --> 00:07:39,299 That’s a huge commitment. 152 00:07:39,299 --> 00:07:41,299 Almost risky, as you’d say. 153 00:07:41,299 --> 00:07:42,299 Yes. 154 00:07:42,299 --> 00:07:44,299 But above all it’s a strategic marker. 155 00:07:44,299 --> 00:07:47,299 It means their absolute priority is flexibility. 156 00:07:47,299 --> 00:07:49,299 Universal access. 157 00:07:49,299 --> 00:07:50,299 The message is clear. 158 00:07:50,299 --> 00:07:53,299 Learning is no longer confined to a desk. 159 00:07:53,299 --> 00:07:56,299 It has to be possible anywhere, anytime. 160 00:07:56,299 --> 00:07:58,299 Without any loss of quality. 161 00:07:58,299 --> 00:08:00,299 That’s a very high standard they’re setting for themselves. 162 00:08:00,299 --> 00:08:03,299 If we step back a bit now, looking at the overall vision, 163 00:08:03,299 --> 00:08:06,299 one detail that isn’t really a detail is multilingual support. 164 00:08:06,299 --> 00:08:08,299 We’re talking about more than 20 countries, 165 00:08:08,299 --> 00:08:11,299 with a special mention for right‑to‑left languages, 166 00:08:11,299 --> 00:08:12,299 like Arabic. 167 00:08:12,299 --> 00:08:14,299 Which, technically, is anything but simple. 168 00:08:14,299 --> 00:08:17,299 It shows that global ambition is built in from the start. 169 00:08:17,299 --> 00:08:19,299 It’s not some distant project. 170 00:08:19,299 --> 00:08:20,299 No. 171 00:08:20,299 --> 00:08:24,299 And that ambition is confirmed when you look at the section about the dashboard, 172 00:08:24,299 --> 00:08:26,299 the admin dashboard. 173 00:08:26,299 --> 00:08:28,299 Of course, there are the basic features, 174 00:08:28,299 --> 00:08:31,299 tracking learners, managing courses. 175 00:08:31,299 --> 00:08:34,299 But they mention two types of reports that stand out. 176 00:08:34,299 --> 00:08:36,299 Analytics reports and... 177 00:08:36,299 --> 00:08:38,299 Ethical use reports. 178 00:08:38,299 --> 00:08:39,299 That’s right. 179 00:08:39,299 --> 00:08:42,299 Now, that last point really caught my attention. 180 00:08:42,299 --> 00:08:46,299 What does an ethical use report look like in practice? 181 00:08:46,299 --> 00:08:48,299 Does it monitor AI bias? 182 00:08:48,299 --> 00:08:50,299 Or if a student is trying to cheat? 183 00:08:50,299 --> 00:08:52,299 That’s an excellent question. 184 00:08:52,299 --> 00:08:55,299 And I think it has to be interpreted from a strategic and commercial angle. 185 00:08:55,299 --> 00:08:56,299 Ah! 186 00:08:56,299 --> 00:09:00,299 The idea is probably not so much to police ethics 187 00:09:00,299 --> 00:09:03,299 as to position themselves as an enterprise‑ready solution, 188 00:09:03,299 --> 00:09:05,299 ready for large accounts. 189 00:09:05,299 --> 00:09:08,299 When you’re talking to a university or a big company, 190 00:09:08,299 --> 00:09:10,299 you’re not just talking to educators. 191 00:09:10,299 --> 00:09:12,299 You’re talking to lawyers, compliance officers. 192 00:09:12,299 --> 00:09:13,299 Right. 193 00:09:13,299 --> 00:09:15,299 People with very strict requirements 194 00:09:15,299 --> 00:09:18,299 on data protection, traceability. 195 00:09:18,299 --> 00:09:19,299 Exactly. 196 00:09:19,299 --> 00:09:21,299 By offering a report on these topics, 197 00:09:21,299 --> 00:09:23,299 Startik is sending them a very clear message. 198 00:09:23,299 --> 00:09:25,299 We understand your constraints. 199 00:09:25,299 --> 00:09:27,299 We are a reliable, responsible partner. 200 00:09:27,299 --> 00:09:29,299 It’s a selling point. 201 00:09:29,299 --> 00:09:31,299 An extremely powerful selling point. 202 00:09:31,299 --> 00:09:34,299 It sets them apart from tools that might be more agile but less structured. 203 00:09:34,299 --> 00:09:38,299 It’s less a moral statement than proof of maturity. 204 00:09:38,299 --> 00:09:39,299 It’s very clever. 205 00:09:39,299 --> 00:09:41,299 If we put all the pieces together, 206 00:09:41,299 --> 00:09:43,299 the picture that emerges 207 00:09:43,299 --> 00:09:46,299 is that of a truly all‑in‑one solution. 208 00:09:46,299 --> 00:09:49,299 We start from the promise of transforming static content 209 00:09:49,299 --> 00:09:51,299 into a dynamic experience. 210 00:09:51,299 --> 00:09:54,299 The engine is an AI that aims to be adaptive 211 00:09:54,299 --> 00:09:56,299 and, above all, grounded in the sources. 212 00:09:56,299 --> 00:09:59,299 Thanks to this RAG‑type architecture. 213 00:09:59,299 --> 00:10:00,299 That’s right. 214 00:10:00,299 --> 00:10:02,299 All of it delivered through an interface designed 215 00:10:02,299 --> 00:10:04,299 to be as smooth as possible, 216 00:10:04,299 --> 00:10:06,299 with a global vision 217 00:10:06,299 --> 00:10:09,299 and a positioning meant to reassure large organizations. 218 00:10:09,299 --> 00:10:13,299 By the way, their final slogan is brutally simple. 219 00:10:13,299 --> 00:10:14,299 Start in 10 seconds. 220 00:10:14,299 --> 00:10:16,299 The summary is perfect. 221 00:10:16,299 --> 00:10:18,299 And all of this leads us to a question, 222 00:10:18,299 --> 00:10:21,299 a final thought, a question to ponder. 223 00:10:21,299 --> 00:10:23,299 I’m listening. 224 00:10:23,299 --> 00:10:26,299 If a tool like this really delivers on its promises, 225 00:10:26,299 --> 00:10:29,299 if it can truly convert any document, 226 00:10:29,299 --> 00:10:33,299 any video into an interactive AI tutor, 227 00:10:33,299 --> 00:10:37,299 then it changes the very definition of what teaching material is. 228 00:10:37,299 --> 00:10:38,299 That’s true. 229 00:10:38,299 --> 00:10:41,299 A research article, the script of a lecture, 230 00:10:41,299 --> 00:10:43,299 an annual report, 231 00:10:43,299 --> 00:10:46,299 all of it becomes potentially the raw material 232 00:10:46,299 --> 00:10:48,299 for personalized learning. 233 00:10:48,299 --> 00:10:51,299 And so the question for trainers shifts. 234 00:10:51,299 --> 00:10:53,299 It’s no longer just 235 00:10:53,299 --> 00:10:56,299 “What content do I need to create to teach this concept?” 236 00:10:56,299 --> 00:10:57,299 But rather 237 00:10:57,299 --> 00:10:59,299 “What conversations, 238 00:10:59,299 --> 00:11:01,299 what explorations can we trigger 239 00:11:01,299 --> 00:11:04,299 from the vast ocean of content 240 00:11:04,299 --> 00:11:06,299 that already exists all around us?” 241 00:11:06,299 --> 00:11:08,299 And that is quite a dizzying 242 00:11:08,299 --> 00:11:10,299 paradigm shift.

Startik, the AI that transforms RAG content

---

General promise and problem addressed

• Transformation in a few minutes of existing content (PDF, audio, video, web pages) into interactive learning experiences
• AI taking on the role of personal tutor, going beyond a simple online course
• Addressing a twofold problem: overwhelmed trainers with “sleeping” content and learners seeking interaction and personalization

Content transformation and pedagogical scenarios

• Versatile sources: audio, video, PDF, Word, web pages, or direct creation in the tool with AI assistance
• Production of rich activities: infinite and adaptive quizzes, AI-guided debates, intellectual simulations
• In debates: the AI adopts a position (sometimes an opposing one) to push the learner to argue based on the sources
• Objective: structure thinking, search for evidence, develop critical thinking rather than “winning” against the machine
• Key limitation: quality of results depends on the quality of the original content (garbage in, garbage out)

Smarter Startik AI: understanding, RAG and adaptation

• Ambition: understand the learner’s intentions, not just keywords
• Automatic search: the AI taps first into the trainer’s documents, then into the web if needed
• Hint of RAG architecture: integration of source excerpts into reasoning to limit hallucinations
• Answers anchored in the course, not in a random Internet
• Continuous progress tracking: adaptation of level, change of explanations (analogies, examples) in case of blockage
• Shift from an AI “that gives the right answer” to an AI “that helps you understand why it is right”
• Trainer-driven personalization: tone, response format, AI personality depending on the nature of the course

User experience and ease of use

• Focus on smoothness: response speed, elimination of anything that breaks the pace
• Advanced support: mathematical formulas, graphs, adaptation to scientific and technical fields
• Systematic display of sources in the AI’s answers to reinforce trust and traceability
• Reduction of “cognitive friction”: the user concentrates on the content, not the platform
• Background audio playback on mobile: integrating learning into daily life (commuting, sports, etc.)
• Promise of an experience more than 99% identical across mobile, tablet and computer: priority on flexibility and universal access

Global vision, multilingual reach and governance

• Support for more than 20 countries, including languages written from right to left (e.g. Arabic)
• Global ambition built in from the design stage, not added afterwards
• Comprehensive dashboard: learner tracking, course management, advanced analytics reports
• Reports on ethical use: meeting compliance requirements (data, traceability) of large companies and universities
• “Enterprise ready” positioning to reassure legal teams and compliance officers as much as educators

Synthesis and paradigm shift for training

• All-in-one solution: transformation of static content into dynamic experience, adaptive AI engine anchored in sources, smooth interface, global vision and robust governance
• Key slogan: “Start in 10 seconds” as a synthesis of the promise of simplicity
• Redefining learning materials: any document (article, conference script, annual report) becomes raw material for personalized learning
• Shifting the question for trainers: from “What content should I create?” to “What conversations and explorations should I trigger from the ocean of existing content?”
• Emergence of a new paradigm where the central challenge becomes orchestrating interactions between learner, AI and content, more than just producing resources