Building AI Agents and RAG Apps Roadmap

Go beyond chat: give a model tools and your own data, and learn how to keep the result reliable and affordable.

9 milestones · ~136 hours total

1. 1

   LLM API basics

   ~10h

   Call a model API, handle streaming, errors, rate limits and keep keys on the server.

   Practice this →
2. 2

   Embeddings and vector search

   ~12h

   Turn text into embeddings, store them in a vector index and retrieve the closest passages.

   Practice this →
3. 3

   Retrieval-augmented generation

   ~20h

   Chunk documents, retrieve, build the prompt, answer with citations and know when retrieval fails.

   Practice this →
4. 4

   Tool use and function calling

   ~14h

   Describe tools with schemas, let the model call them, validate arguments and return results safely.

   Practice this →
5. 5

   Agent loops

   ~18h

   Plan, act, observe: build a small agent with a step limit, memory and clear stop conditions.

   Practice this →
6. 6

   Evaluation

   ~14h

   Create test questions with expected answers, measure retrieval and answer quality and watch for regressions.

   Practice this →
7. 7

   Cost, latency and caching

   ~8h

   Pick model sizes, cache repeated work, trim context and set budgets and timeouts.

   Practice this →
8. 8

   Security and guardrails

   ~10h

   Prompt injection, data leaks, permission boundaries for tools and human approval for risky actions.

   Practice this →
9. 9

   Capstone project

   ~30h

   Ship a document question-answering or support assistant for a real use case, with tests and a short write-up.

   Practice this →