Overview
This project applies our engineering approach outside regulated industries, in a commercial context where correctness, consistency, and production quality still matter.
The outcome is an AI-assisted system for generating educational children’s flashcards intended for physical printing and sale, not digital experimentation.
Although this is not a regulated financial product, the project imposed strict constraints comparable to those found in regulated environments.
Problem context
Educational products for children carry implicit quality and trust requirements:
- Content must be factually correct and age-appropriate
- Information should be traceable to credible sources
- Outputs must be consistent across large production runs
- Errors discovered post-printing are costly and irreversible
The challenge was to design a system that could assist with content generation without sacrificing evidence, reviewability, or production quality.
Engineering approach
The system was designed as a controlled, multi-stage pipeline rather than an end-to-end generative tool.
Key principles included:
- Every card defined against an explicit schema and validated before anything is generated
- Generation split into small, single-purpose stages rather than one end-to-end prompt
- Text and imagery produced separately, each reviewed on its own terms
- Content and artwork kept under version control, so every printed card traces back to a reviewed revision
- Press-ready output (PDF/X-4) built reproducibly from source, not assembled by hand
The general-purpose parts of the pipeline have since been published as open source by Release.art: texmodel, texpkg, and proofpress. See Release.art for the full list.
Quality assurance and safeguards
To avoid the risks commonly associated with generative content, the system includes explicit safeguards:
- Separation between content generation and approval
- Source attribution alongside generated facts
- Manual sign-off before inclusion in print runs
- Clear constraints on tone, vocabulary, and scope
- Pre-flight checks for print correctness and colour handling
Why this project matters
This work sits outside regulated sectors, but it follows the same engineering principles as our core services:
- Explicit constraints over implicit behaviour
- Evidence and provenance over plausibility
- Reviewable outputs over opaque automation
- Quality gates at every irreversible step
It shows that disciplined automation can be used responsibly in commercial, quality-critical products as well as compliance systems.
Scope and positioning
This project is presented as an applied example, not a standard product offering.
We take on this type of work selectively, where the requirements fit our principles on evidence, governance, and defensible outputs.