AI & Generative AI Roadmap

From how language models work to shipping a safe, grounded AI feature — the practical path for a developer.

9 milestones · ~113 hours total

1. 1

   How LLMs work

   ~10h

   Tokens, embeddings, attention in plain language, pre-training vs fine-tuning, context windows, and why models hallucinate.

   Practice this →
2. 2

   Prompting well

   ~12h

   System prompts, few-shot examples, structured output, chain-of-thought, and testing prompts like you test code.

3. 3

   Sampling and decoding

   ~6h

   Temperature, top-k, top-p; how they change output and when to use which. Implement softmax with temperature yourself.

   Practice this →
4. 4

   Embeddings and semantic search

   ~12h

   Create embeddings, compare with cosine similarity, build a small search over your own notes.

   Practice this →
5. 5

   Retrieval-Augmented Generation

   ~20h

   Chunking, indexing in a vector store, retrieval, grounding answers with citations, and evaluating quality.

   Practice this →
6. 6

   Tools and agents

   ~15h

   Function calling, multi-step workflows, and when an agent is overkill. Permission design and human confirmation.

7. 7

   Safety and evaluation

   ~12h

   Prompt injection, data leakage, bias, privacy, and building a small evaluation set to catch regressions.

   Practice this →
8. 8

   Ship a project

   ~25h

   Build and deploy one useful AI feature end to end with a README that states its limits honestly.

9. 9

   Take a rated AI & Generative AI test

   ~1h

   Take a timed rated test to earn your first verified AI & Generative AI rating — the number recruiters see.

   Practice this →