AI Training Foundations
What training a model actually means, and how to set one up to succeed.
- What it means to train an AI model
- Define the task and its success criteria
- Choose a base model
- Curate data that teaches
Learn to train AI, then go and train it.
5 free courses and 24 lessons on how models and agents are trained, evaluated and shipped in 2026, from your first prompt to reasoning models, agent evals and the EU AI Act. No account needed to read.
Prompt Tuning, Fine-Tuning & Reasoning
From a hand-written prompt to an optimised one, and when to go further.
- Prompting fundamentals
- Automatic prompt optimisation
- Fine-tuning efficiently with LoRA and QLoRA
- Learning from preferences: RLHF and DPO
- Reasoning models and verifiable rewards
The Economics of Model Training
What a training run takes in compute, memory, hardware and money.
- How much compute a training run takes
- Memory first: what hardware can train what
- What training costs at each model size
- Rent, buy or don't train at all
Evaluation & Trust
If you haven't measured it, you haven't trained it.
- Test sets that tell the truth
- LLM-as-a-judge, and its biases
- Measuring hallucination and bias
- Evaluating agents: trials, transcripts and pass^k
- Latency, trust scores and the ship decision
Agents in Production
Tools, context, security and feedback for agents that run unattended.
- Anatomy of an agent
- Context engineering
- Red-teaming and agent security
- Deploy, monitor and learn from feedback
- Connecting agents with the API and MCP
- Governance and risk in 2026
Every claim is cited
Lessons link to the papers, specs and standards behind them, so you can check the source.
Practise in the dojo
Each lesson ends with a hands-on exercise in Kojen: a kata, a Forge run, an evaluation or a deployment.
Check what you learned
A short quiz closes every lesson. Score 70 or more to complete it, the same bar as a kata.
Ready to practise?
Kojen is free to start, then pay as you go. Create an account to run the exercises: katas, Forge runs, evaluations, simulations and live deployments.