A Self-Built Coach-LLM in the Classroom: From Idea to Practice

A Self-Built Coach-LLM in the Classroom: From Idea to Practice

Imagine always having a positive, cheerful, and concise coach at your side as a teacher, someone who helps motivate students, asks open-ended questions, and points them to the right resources. With a self-trained Large Language Model (LLM), this can become a reality! In this blog, you’ll learn how to create your own LLM as a digital coach for your students.

1. What Is a Coach-LLM?

A Coach-LLM is an AI model you train using your own documentation, lesson materials, and coaching guidelines. The model is designed to, in a cheerful way and using clear child-friendly language, support your students with design processes, reflections, and other assignments. The goal is not to provide ready-made answers, but to encourage students to arrive at insights themselves.

By using the right “prompt” (instructions) and a dedicated dataset (such as your lesson documents), the model can respond to each student’s question with coaching feedback. Think about:

Asking open-ended questions (“What have you already investigated?”)
Stimulating self-reflection (“Which steps have you already taken, and how did that go?”)
Referring to sources (for example: “Have you looked in 2024 – Reflectie opdracht.docx for tips on describing your learning process?”)
Envision the questions your students might ask, pinpoint the documents they’d need, and outline exactly how your CustomGPT should guide them.

2. Preparation: Collect the Right Documents and Sources

Begin by compiling a dataset that will “feed” your model. The “Coachend Reageren” document references various sources that form the basis of your Coach-LLM. Examples include:

Reflection paper: Guides students in analyzing their own learning process.
Coaching Guideline: General tips for coaching behavior, such as listening and asking open questions.
Rubrics Beoordeling: A set of criteria that helps students assess their own level and see what’s required.
How to make a inforgraphic: Tips for creating professional infographics.
How to create a Essay in Word: Handouts to practice and improve Word skills.
How to create To Do's in Scrum: Practical Scrum applications for project-based lessons.
How to make a presentation: Guidance on creating and delivering presentations.
How to setup your essay: Assistance with writing a design report.

Important: The guidelines emphasize not providing solutions but always responding in a coaching manner and referring students to these documents.

3. How to Build Your Own Coach-LLM

Step 1: Collect and Clean Your Data

Gather your documents in a single folder or repository.
Remove superfluous information so only relevant text remains.
Highlight the essence of your coaching approach (such as example questions and the tone you want to use).

Step 2: Set Up a Training Environment

Choose a platform: for example, OpenAI’s Custom GPT.
Create an account or install the software, depending on whether you want to train online or locally.

Step 3: Fine-Tune with Your Own Guidelines

Use your collected documents as a “prompt” or as training data.
Define clear prompt rules—for instance:
- “Always write briefly, clearly, and in child-friendly language.”
- “Ask open-ended questions and encourage reflection; refer to the correct document.”
- “Never provide the direct answer, only coaching questions and tips.”
- My custom GPT 'quick' made guideline below:

Step 4: Test and Validate

Pose various student questions to check if the LLM responds consistently.
Pay attention to tone: is it always positive and encouraging?
Verify that the model refers to the right sources (for example: “Have you looked at Rubrics Beoordeling to see what you’re missing?”).

4. Benefits for You and Your Students

More personalized attention: While you help other students, the Coach-LLM can answer questions from those who are waiting.
Consistent coaching attitude: The model always remains cheerful, positive, and uses child-friendly language, no more risk of a terse response on a busy day.
Increased student autonomy: Because it prompts them with open questions, students learn to seek answers themselves and consult resources on their own.

Demonstration video in Dutch

If you want to try out the bot, click on the image:

Conclusion

As AI-powered coaching tools become more simpel, the conversation is shifting beyond simply how to build a custom LLM. While the technological barriers to training or fine-tuning a language model have significantly lowered, the deeper and more critical questions revolve around the wider implications for our education ecosystem:

Digital Literacy: The presence of AI coaches in the classroom challenges students and teachers alike to develop higher levels of digital literacy. Students must learn not only how to use these tools, but also how to critically assess and interpret AI-generated feedback. This shift underscores the need for structured digital-literacy curricula that go beyond basic computer skills and address ethical, privacy, and evaluative dimensions of AI.
Curriculum Development: As AI-driven assistance becomes more commonplace, curriculum designers will need to adapt their materials and approaches. Rather than focusing solely on content knowledge, new curricula should encourage critical thinking about AI tools, when to rely on them, when to question them, and how to use them effectively. The integration of these skills throughout lesson plans could help prepare students for an ever-evolving digital landscape.
Positives of AI Coaches
- Personalization: Students receive near-instant, tailored feedback, supporting diverse learning styles.
- Scalability: AI coaches can assist large cohorts of students simultaneously, freeing up teachers to address complex individual needs.
- Self-Paced Learning: Learners can engage with AI-driven feedback at their own speed, enhancing autonomy and ownership of the learning process.
Negatives and Potential Pitfalls
- Overreliance on Automation: Students (and teachers) may lean too heavily on AI for answers, risking the erosion of essential problem-solving skills.
- Bias and Quality of Feedback: AI models can inadvertently reinforce biases or provide incomplete advice if not carefully reviewed.
- Digital Divide: Not all schools or communities have equitable access to the technology, exacerbating existing educational gaps.
The Evolving Role of the Teacher: In an environment where AI coaches can take on certain tutoring or feedback tasks, the teacher’s role becomes more multifaceted and human-centered. Rather than simply delivering content, teachers will increasingly function as:
- Facilitators of Critical Thinking: Helping students evaluate AI feedback, question assumptions, and develop deeper problem-solving strategies.
- Curators of Learning Paths: Selecting or designing AI-assisted materials that fit individual students’ progress and needs.
- Ethical Guides: Ensuring that discussions around AI usage remain centered on responsibility, digital citizenship, and empathy.

Practical Considerations for Successful AI Integration

Set clear expectations: Students need to understand that the LLM is not a homework machine but a tool to help them think.
Stay involved: Regularly check whether the LLM is giving correct advice and not accidentally revealing solutions.
Respect privacy: Don’t collect or process any sensitive student data in the model.
Continue evaluating: Ask students for feedback. Do they find the model truly helpful for reflection? What could be improved?