Personal LLM Engineering Programme

This programme studies two connected machines:

1. the mathematical machine that turns text, probabilities, errors, and parameter updates into a language model;
2. the engineering organisation that turns data, experiments, accelerators, distributed systems, and people into a reliable training campaign.

It is designed for an experienced software engineer. Its goal is architectural and theoretical understanding, not preparation for operating a frontier model training run.

The three courses

Course I — How Language Models Learn

From tokens and tensors to loss, gradients, attention, and Transformer blocks.

Course II — Training as Data and Experimental Engineering

From raw corpora and evaluation to ablations, scaling, post-training, and the hill-climbing development loop.

Course III — Training Systems at Scale

From accelerator kernels and collective communication to orchestration, checkpointing, observability, goodput, and team responsibilities.

Current production state

The curriculum map is provisional. Unit 1 is ready for learner review. Units 2 and 3 have research packets but remain planned, allowing the real study experience of Unit 1 to calibrate their authoring and the rest of the programme.

1. The Model Lifecycle
2. Tokens, IDs, Vectors, and Tensors
3. Next-Token Prediction