From AI Fundamentals to LLMs: Build Your Own Language Model from Scratch

This 4-day training (28 hours) offers a progressive and structured learning path, from discovering artificial intelligence to the full implementation of a language model. The first day lays the foundations: history of AI, key concepts, real-world use cases, and an introduction to machine learning with a first prediction project (linear regression on a Kaggle dataset). The second day dives into deep learning: neural networks, gradient descent, backpropagation, and training with PyTorch on the MNIST dataset. A hands-on workshop introduces no-code AI tools through the creation of a fictional startup. The last two days are dedicated to building a complete transformer step by step: tokenization, embeddings, attention mechanism, followed by the full implementation of the GPT-2 architecture and training on a text corpus. The training concludes with fine-tuning, inference, LLM evaluation, and an overview of enterprise applications: RAG, agents, quantization, vLLM, and Model Context Protocol (MCP).

Day 1: Demystifying AI and Introduction to Machine Learning 

Objective / Output : 

• A summarized overview of the history of AI 
• Understanding of how AI works 
• Knowledge of key terms; Bonus: Glossary 
• Knowledge of standard AI tools; Bonus: our recommendations 
• Ideas on where AI can help you

Day 2: Introduction to Deep Learning and Hands-on No-Code Workshop 

Objective / Output : 

• Understand neural networks, gradient descent, and backpropagation 
• Train a model on the MNIST dataset with PyTorch and discover no-code AI tools 

Day 3: Transformer Architecture and Building GPT 

Objective / Output :

• Understand tokenization, embeddings, and the attention mechanism 
• Build the complete transformer block (LayerNorm, Dropout, Feed Forward, Shortcut connections) 
• Master text generation: greedy decoding, temperature, and top-k sampling 

Day 4: Coding GPT-2 from Scratch and LLMs in the Enterprise 

Objective / Output : 

• Implement the complete GPT-2 architecture in Python 
• Train the model on a text corpus 
• Apply fine-tuning (SFT) to turn a base model into an assistant 
• Understand inference, quantization, and vLLM 
• Discover RAG and the Model Context Protocol (MCP) 
• Evaluate an LLM (MMLU, benchmarks, LLM-as-judge).