Kinnu Builds a Knowledge Graph With AI-Generated Learning Paths

01 /The Concept

Kinnu raised £5.1M ($6.5M) at seed stage on a proposition that initially sounds either visionary or hubristic: build a structured, personalized learning system covering every major knowledge domain, complete with curated content, adaptive learning paths, and spaced repetition – and eventually automate most of the content creation with AI.

The platform is not a course library. It is a knowledge graph: interconnected articles, each readable or listenable in 5–15 minutes, organized by topic and linked through learning pathways called "pathways" that map logical progressions from starting knowledge level to target understanding. A data analyst exploring machine learning, a marketing manager learning financial modeling, and a statistician picking up a new programming language would each follow different pathways through overlapping content – the same articles appearing in multiple sequences depending on where the learner starts.

Content is currently produced through a human-AI hybrid process. A subject expert builds a skeleton outline. An AI model generates the full text, quiz questions, and retention exercises. An external domain expert edits for accuracy and corrects hallucinations. An internal editor adds AI-generated illustrations and AI-synthesized audio. The full cycle for one pathway takes under 10 hours of human labor. The eventual goal is near-full automation with minimal editorial review – and ultimately, a system where a learner can say "I want to understand data analysis in five hours and I have a light statistics background" and Kinnu generates a personalized curriculum on demand.

Retention is built in: spaced repetition prompts resurface material before it fades, using a mix of simple reminders, multiple-choice quizzes, and engagement mechanics designed to sustain the learning habit rather than just the initial session.

02 /Why It Matters

Kinnu's actual innovation is less about the product experience and more about the production model. A company like Go1 ($413.7M raised) aggregates and licenses existing third-party content. Odilo ($84.9M) and Obrizium ($12.4M) build learning path infrastructure on top of it. The expensive assumption in each case is that good content has to be sourced, licensed, or commissioned.

Kinnu's bet is that AI has made original content generation cheap enough to eliminate that assumption entirely – and that the real value in the stack is the pathway architecture and the personalization layer, not the content itself. Wikipedia has 7 million articles in multiple languages on every conceivable topic; the learning value is not in the existence of those articles but in knowing which ones to read in which order for a specific goal. That is the job Kinnu is hiring itself for.

The content volume problem is also less daunting than it appears. Wikipedia's long tail includes enormous numbers of articles with very low view counts – most knowledge consumption is concentrated in a relatively small number of popular topics and progressions. Kinnu's internal estimate of 1,000–5,000 pathways to cover the majority of user demand is not as optimistic as it sounds when viewed through the lens of how app downloads distribute: only 4.6% of App Store apps are downloaded more than 100,000 times. Most of the addressable learning demand is concentrated in a tractable set of subjects.

The more pointed observation is about the current state of online education. A meaningful portion of course creators are already using AI to generate curriculum outlines and lesson text – and selling the result as original expert content. At the same time, sophisticated learners are using the same AI tools to generate custom courses for themselves. Kinnu is essentially trying to productize the better version of that second behavior: structured, fact-checked, maintained over time, and embedded with actual retention mechanics rather than just content delivery.

03 /Opportunities

The clearest opportunity is in self-directed learning infrastructure – platforms that let learners generate effective study paths from their own specifications rather than choosing from a pre-built course catalog. The demand for this exists and is growing: the question is how much editorial scaffolding is required to make AI-generated content trustworthy enough for professional or high-stakes learning contexts.

Kinnu's current approach – building a proprietary content base with expert validation – is defensible but expensive. A lighter version of the same concept could focus on pathway architecture alone: a system that ingests a learner's goal and knowledge state, maps it to a sequence of high-quality existing resources (curated from public or licensed content), and applies spaced repetition on top. This avoids the content-creation bottleneck while still providing the personalization value that Kinnu is targeting.

The timing argument is real and worth taking seriously. AI capabilities for content generation, hallucination correction, and curriculum personalization are improving faster than most traditional education providers are adapting. The companies that build learning infrastructure now – even imperfect early versions – will accumulate learner data that compounds into a durable advantage. The window for being an early infrastructure provider in AI-native education is open, but probably not for much longer. Entering with a focused vertical or subject domain is lower-risk than trying to build a universal system from the start.