SMART Technologies

Teachers want interactive lessons. They do not have time to build them. Translating a casual brief into a clean, classroom-ready activity (with prompts, drag-and-drop interactions, sensible visuals) is days of work, not minutes, and it is happening in a normal prep window that is already full.

Anything that closed that gap also had to clear a higher bar than a consumer LLM toy: content safety, predictable latency under load, and graceful degradation when one provider was unavailable.

We shipped a generation pipeline. Claude 3 Sonnet decomposes the brief and drafts the activity. A Rust Actix-Web service moderates, spell-checks, and sanitises the output, with automated filters and a human-in-the-loop review path. A Next.js frontend streams the activity to the browser as it is generated, so teachers see and edit it live rather than waiting on a spinner.

The model layer was deliberately provider-agnostic, with fallbacks to OpenAI, Anthropic, Meta, and xAI. The system embeds directly into SMART Boards and LMSs, so the output lands in the surface teachers and students already use, instead of yet another tab.

Across an eight-month pilot in five North American schools, the system generated more than 300 classroom activities, reaching 150+ students. Prep that used to be measured in hours collapsed to minutes, and the activities were good enough that teachers chose to use them.