Research, engineering,
design.
Ideas behind the products. What we are building, why we built it this way, and what we learned.
Why retention is the only metric that matters
Engagement rates tell you whether learners showed up. Completion rates tell you whether they stayed. Neither tells you whether anything was learned. We built Arconite around the only metric that does.
Routing models by cognitive task
Foundation models are generalists. Learning requires specialists. Inside Talos, every inference request is classified by cognitive type before a model is selected — and the routing logic is the most important code we have written.
Designing calm interfaces for learning
Most software competes for attention. Learning software should protect it. The design language behind Gripho and Skemax starts with a simple principle: every element earns its presence or it does not exist.
Memory anchors: spaced recall inside conversation
Spaced repetition works. Flashcard apps are unpopular. The gap between those two facts is a design problem, and we built the solution into the conversational layer of Gripho.
Skemas: treating courses as systems
A course is not a document. It is a system with inputs, outputs, feedback loops, and failure modes. Skemax was built on that premise, and the skema is the data structure that makes it real.
Inside Talos: one intelligence layer, many models
Talos is not a single model. It is an orchestration architecture that coordinates foundation models, specialised models, retrieval systems, and validation layers into one coherent intelligence layer for every Arconite product.