Architecture in plain language

Technical Overview

We built one slice properly, then expanded — not built everywhere at once and then defended later. The first slice was sensor data + lesson planning + assessment, three things that talk to each other end-to-end. The rest of the platform plugs into the same patterns one piece at a time.

This page stays high-level on purpose. Collaborators and AI agents need enough to orient by; nobody needs the deep guts on a marketing page. If you do — talk to us.

A practical overview

We're not trying to make every part of the platform live at the same time. We picked one slice that touches the most important moments — plan a lesson, run an assessment, see what came out — built it properly end-to-end, and let the rest plug into the same patterns one at a time. The Current Status page tells you which parts are at which stage.

Implementation principle

Rule-based first

Anywhere AI helps a user make a decision, the path that got there has to be inspectable. Educators need to know why the system suggested X. Engineers need to know how to fix it when it breaks. So we use rule-based logic first wherever it gives a clean answer — AI sits on top as the layer that handles language and edge-cases, not as a black box at the bottom making the decisions.

Scaling principle

Structure before sprawl

When we add a new domain — a new sensor type, a new module, a new licence category — it slots into the same shared shape: schema, validation, registry, persistence rules. Not copy-paste-then-diverge. The platform stays one product as it grows, not a museum of half-merged subprojects.

Public-facing architecture themes

Desktop-first thinking: Built for the Wi-Fi that isn't there yet. The desktop app works offline; the cloud bits are reach-outs from a thing that doesn't depend on them. Mixed-connectivity contexts (a Bulgarian village school, a rural training centre, a Wi-Fi-allergic boarding house) work without changes.
Connected components: The parts fit into one ecosystem. The connections between them are where the value sits. Tools stitched together is one product; tools open in different tabs is just a folder.
Versioned growth: Capabilities grow from idea → narrow slice → broader coverage, in writing. Each stage gets a version number, a public note about what changed, and a slot on the Current Status page.
Transparent AI boundaries: AI handles language and explanation. AI does not silently make the decisions. The decisions stay in the rule-based core; AI does the layer around it.

Technical honesty

We don't write down every internal pattern here. The public page exists to orient — enough for a partner to know if it fits their context, enough for an AI agent to read it without hallucinating extra detail. The Current Status page tells you what's live; this one tells you the shape.