Building an Adaptive Knowledge Graph

A build retrospective: modeling a whole Arabic-learning curriculum as one knowledge graph, layering a per-child learner model on top, and generating every surface — app, web, printables, books, marketing — from that single source. The real system behind Amal, live for 95,000+ families.

1. 01The Problem With Flat CurriculaA build retrospective on the adaptive knowledge graph behind Amal, a kids' Arabic-learning app used by 95,000+ families. • Day 1 sets the stage: why a plain list of lessons quietly fails every child, and the bet that fixed it — model the whole curriculum as a graph.7 sections
2. 02Modeling the Curriculum as a GraphA build retrospective on the adaptive knowledge graph behind Amal, a kids' Arabic-learning app used by 95,000+ families. • Day 2 makes the graph concrete: the exact tables that store the concepts, the prerequisite edges between them, and the pictures and audio each concept carries.7 sections
3. 03The Learner OverlayA build retrospective on the adaptive knowledge graph behind Amal, a kids' Arabic-learning app used by 95,000+ families. • Day 3 adds the second half of the system: a per-child layer on top of the shared graph that records what each child knows and predicts what they are about to forget.7 sections
4. 04The Learner FrontierA build retrospective on the adaptive knowledge graph behind Amal, a kids' Arabic-learning app used by 95,000+ families. • Day 4 is the heart of the engine: a single pure function that takes the shared graph plus one child's overlay and returns exactly the concepts that child is ready to learn next.7 sections
5. 05The Graph Tree BuilderA build retrospective on the adaptive knowledge graph behind Amal, a kids' Arabic-learning app used by 95,000+ families. • Day 5 turns the frontier into something a child actually sees: a per-child, realtime subject tree — two children get two different trees from one graph.6 sections
6. 06Always-Generative BytesA build retrospective on the adaptive knowledge graph behind Amal, a kids' Arabic-learning app used by 95,000+ families. • Day 6 fills in the last engine stage: generating the actual lesson from a concept on demand, instead of hand-authoring every lesson in advance.6 sections
7. 07One Graph, Every SurfaceA build retrospective on the adaptive knowledge graph behind Amal, a kids' Arabic-learning app used by 95,000+ families. • Day 7 is the payoff: one read seam over the graph lets a single concept become an app lesson, a web page, a printable, a book listing, a course, and a marketing card — at ~zero marginal cost per surface.6 sections
8. 08Shipping SafelyA build retrospective on the adaptive knowledge graph behind Amal, a kids' Arabic-learning app used by 95,000+ families. • Day 8 is the engineering discipline: how to drop a brand-new engine underneath a live app used by tens of thousands of families without a single child seeing a broken screen.7 sections
9. 09Proving AdaptivityA build retrospective on the adaptive knowledge graph behind Amal, a kids' Arabic-learning app used by 95,000+ families. • Day 9 is the honest chapter: the moment the adaptivity signal turned out to be lying — every child looked identically "mastered" — and the fix that made mastery mean something again.6 sections
10. 10Results and What's NextA build retrospective on the adaptive knowledge graph behind Amal, a kids' Arabic-learning app used by 95,000+ families. • Day 10 lands the story: what the finished system does in the real world, the map it gives parents, and where it goes next.6 sections