Rapid AI-assisted prototyping as a working method for learning and development products. One of several threads in this practice. See the systems-change thesis.
Starting from a single inference API, I built a series of small products to find out what the technology could actually do in learning and development. Each one shipped, went in front of real users, and then either earned its place or was retired.
The method is the same one I apply to any complex problem. Run experiments that are small enough to be cheap and reversible, watch what happens with real users, and put the effort behind whatever demonstrates value. It is the working version of the approach described in designing an organisation that learns with AI.
The foundation
The Lumination AI API brings several capabilities under one authentication layer: conversational AI with page and message context, document and image understanding, and contextual tutoring threads. Responses come back structured, with token counts and credit usage, so analytics, rate limiting and cost tracking are available from the first build.
The same endpoint powers a floating chatbot on a website, a step-by-step maths solver for students, and a leadership decision simulator. What separates those three products is prompt design, context handling and interface, which turned out to be the most useful lesson of the whole exercise.
What got built
- Leadership decision simulator. Users set a role, a challenge type and a difficulty level. The API generates a realistic workplace scenario with response options, then returns a coaching debrief with scoring once the user decides. This tested whether the model could produce structured, parseable output rather than free text.
- AI coach chatbot. Upload PDFs, images or written notes as context, then hold a conversation against that material. A test of multimodal input handling.
- PDF quiz generator. A document goes in, a multiple-choice quiz with per-answer explanations comes out.
- Live transcription for speech and language therapy. Built in four hours one Saturday for a therapist working with a hearing-impaired patient, where the available apps returned delayed blocks of text that made natural conversation impossible. It did real-time transcription with colour-coded speaker identification. The tool served its purpose and has since been retired, which is the expected end for most experiments.
The surviving experiments run at ai.azvai.com, built with Astro and vanilla JavaScript, with every API call server-side.
What reached production
Lumination Chatbot is a WordPress plugin running on live sites. Its distinguishing feature is page-context awareness: the plugin extracts the current page content server-side and passes it to the API, so the assistant answers questions about the page the visitor is actually reading. Conversations, token usage and costs are tracked in an admin dashboard.
Lumination Homework Helper takes a photo of a maths problem and returns a step-by-step solution with proper mathematical notation. It chains the extraction and conversation endpoints, renders LaTeX reliably inside WordPress, and ships with rate limiting, file validation and a full privacy implementation for GDPR.
Drop your file here or click to browse
Supports PNG, JPEG, PDF (max 10MB)
You can also paste images with Ctrl+V
Live demo. Upload a photo of a maths problem or type one in, and watch it work through the solution.
Building the thing is often the cheapest way to find out whether the idea holds. A prototype in front of real users answers questions a specification never will, and the failures cost hours instead of quarters.
Why it fits this practice
Reusable patterns emerged across the products, covering analytics, rate limiting, response handling and structured prompting, and each one transferred to the next build. That accumulation is what makes the pace sustainable, and it is the same operating model described in what an AI-augmented consultant actually does.
Related work in this thread: the TutorAI learning platform. If you want prototypes or integrations built for your own systems at this pace, our services page explains how we engage.