The teacher only had so many hours.
Helping an independent IT trainer make room for more learners, with a small GPT prototype and a learning platform that keeps support in step with each trainee.
The Master ChefCode Cooks / From the kitchen
The training itself was only part of the work. A lot of it happened after the session ended.
Our client had retired from an executive partner role at a well-regarded consultancy. He had been running training for IT leaders for a while and wanted to reach more people. The courses already existed. So did the experience people were coming to him for.
The difficulty was the work around each course. Follow-up questions, exercises people worked on at home, exams, and the explanations someone needed when a lesson had not quite settled. Each new trainee brought more of that work back to the same person.
Hiring more tutors would have changed something important about the academy. People were paying to learn from him, with his experience and judgment behind the teaching. He wanted to keep that relationship as the number of learners grew.
That gave us a fairly specific question to work on: how could the academy support more people between sessions when its teacher still had the same number of hours?
What information consumes is rather obvious: it consumes the attention of its recipients.
He came to us looking for help with automation. We started with the part that seemed most useful to try: an agent that could work with his course materials and help trainees with their questions and exercises.
The first version was a very small GPT prototype. We gave it his existing course materials and used that as the starting point for an experiment. Could it give useful help with the things he was teaching? Would the answers make sense to the person who knew the subject and had put the courses together?
We spent roughly five to ten hours getting the idea into a form he could try. That was enough work to put the important question in front of him while the experiment was still small.
Then we handed it over for him to play with. He could ask questions, try it against his material, and see what came back. He had the experience to judge those answers. Giving him something to use made the conversation concrete.
We went through a couple of iterations together. Once he was happy with the results, we had enough confidence to start building the platform around it. His reaction to the prototype helped us decide what was worth taking further.
Therefore, the most important function that the software builder performs for the client is the iterative extraction and refinement of the product requirements.
That became a small learning management system. A trainee pays for a course and receives an access link. Inside, they have a place to work with the material and get support between sessions. We wanted the platform to fit around the teaching he was already doing.
Each trainee has their own knowledge base. In ordinary language, that is the collection of course materials made available to that person so far. The agent loads the knowledge base for the trainee it is helping, so those materials provide the context for questions and exercises.
After each session, the trainee’s knowledge base is updated with the new material. The next time they need help, the agent has that material available too. The collection grows as the course moves forward.
That detail mattered more than it might first appear. A course has an order. The teacher introduces an idea, works through it, and builds on it in a later session. We wanted the material supplied to the agent to follow that same progression.
Imagine two trainees on the same course who started at different times. One has attended the first session. The other has reached the third. They are studying the same subject, but the materials available to help each of them should reflect where they have actually got to.
- After session 01Session 01
material - After session 02Sessions 01 + 02
material - After session 03Sessions 01 + 02 + 03
material
A question can have several reasonable answers. The useful one depends partly on what has already been explained. If a reply brings in ideas from a later session, the trainee has to make sense of those as well as the exercise in front of them. It can make a small point of confusion harder to work through.
We gave the knowledge base a job in that interaction: keep the course material provided to the agent in step with the learner. That let us control which parts of the course were supplied at each stage. The support could build on the sessions the trainee had reached.
The same structure also made room for people taking more than one course. A trainee could subscribe to different courses and build up their collection of material over time. Their knowledge base could grow across those courses as new sessions added to it.
That was useful for an academy built around one teacher. Returning for another course could add to what a trainee already had access to. The platform could keep track of that developing collection as the person continued learning with him.
We ended up with a small set of relationships to build around: who was asking, which courses they were taking, and how far they had got. Those relationships determined which materials the agent loaded.
- The traineeTheir courses and progress.
- Their knowledge baseCourse material released to them so far.
- The agentHelp with questions and exercises.
Someone using the platform would see a question and a reply. Behind that exchange was a decision about which teaching material should be available for that person. Those details are easy to miss when the first description of a project is to add an agent to a course.
The exercises mattered in that thinking. A trainee is trying to use an idea and has reached a point they cannot get past. Help needs to make sense from there. The material they have already worked through gives an explanation somewhere to begin.
Thinking is as much an individual matter as is the digestion of food.
Looking back, the small prototype was a useful decision. Five to ten hours gave us something we could discuss with the person whose judgment mattered most to the project. The iterations that followed happened around answers he could actually read and consider.
Building around one experienced person changes what growth has to look like. Some of the value is in the way they explain something or the order in which they introduce it. Those choices need to carry through into the parts of the experience that happen between sessions.
The point was to make room for more learners while keeping his teaching at the center. The courses, their sequence, and the material people worked through still came from him. We built a way for that material to keep being useful between sessions.
For a trainee, that could be a small interaction: an exercise, a question, and enough explanation to take another step. A lot of the thought went into making sure the help began where their last lesson had left them.
