Skip to content
wannacode

AI for developers: Java

The course is the same for all four languages, and the code here is in Java, with no libraries, so you can see every number. The first part is how a model learns, the second is apps with language models.

Module 0

Welcome

1 lesson · the two parts of the course, why the code has no libraries

  1. Welcome

Module 1

How a model learns

6 lessons · regression, gradient descent, classification, evaluation

  1. What machine learning is
  2. Linear regression
  3. Gradient descent
  4. Classification: fraud or not
  5. Evaluation: accuracy lies
  6. Overfitting: the model that memorises

Module 2

A neural network from the inside

5 lessons · neuron, layers, backprop, training

  1. The neuron
  2. Layers
  3. Backpropagation
  4. Training from start to finish
  5. The digits on the receipt

Module 3

How a language model works

5 lessons · tokens, the next token, vectors, context

  1. Text in pieces: tokens
  2. Predicting the next token
  3. Embeddings: meaning as vectors
  4. The context window
  5. The limits of the model

Module 4

Apps with LLMs

6 lessons · APIs, prompts, structured output, tools, cost

  1. Calling a model through an API
  2. The prompt is code
  3. Structured output: JSON by a schema
  4. Tool use: the model asks, the program acts
  5. Evaluation: a test set for the assistant
  6. Cost and latency: budget, cache, a smaller model

Module 5

Semantic search and RAG

4 lessons · document vectors, chunking, answers with a source

  1. Semantic search: the k nearest
  2. Chunking documents
  3. RAG: answers from the documents
  4. Evaluating RAG: does it find it, does it stand on it

Module 6

Agents

3 lessons · the agent loop, prompt injection, guardrails

  1. The agent loop: plan, action, observation
  2. Prompt injection: the email that gives orders
  3. Guardrails: checks before and after the model

Module 7

Project: the support assistant

1 lesson · RAG, a tool, a test set and protection in one

  1. The support assistant

Practice

Tasks

Every task from the lessons in one place, by module.

Made by Ivo Gergov as part of Evo Tech's campaign for accessible programming education ·