Building AI applications with Python

About this track

You can write Python; now learn to build software on top of a language model. This track covers calling a model from code, forcing output into a schema your program can rely on, giving the model tools it can call, the agent loop, retrieval-augmented generation over your own documents, evaluating whether a change actually helped, and controlling cost and latency in production. It is deliberately opinionated about the failure modes, because the gap between a demo that works once and a system that works every day is where most AI projects die. Take Python data analysis first if you have never used pip or a virtual environment.

Every lesson is written and hosted here on Yanjye — you never leave the site. Work through them in order, then sit the exam to earn your certificate.

Create a free account or log in to track your progress and earn the certificate.

Lessons

1. 1
   From chat window to API call

   What changes when the model is inside your program instead of a browser tab.

   ~18 min
2. 2
   Your first program that calls a model

   Client, request, blocks, usage — and reading the response properly.

   ~22 min
3. 3
   Prompts in code: system design, not wording

   Treating the prompt as a versioned, testable part of your program.

   ~20 min
4. 4
   Getting JSON your program can rely on

   Schema-constrained output instead of parsing prose and hoping.

   ~22 min account needed
5. 5
   Giving the model tools it can call

   Function calling: how the model reaches your database, your API, the real world.

   ~25 min account needed
6. 6
   The agent loop, and how to keep it under control

   Multi-step autonomy, budgets, stopping conditions and failure recovery.

   ~22 min account needed
7. 7
   Embeddings: turning meaning into numbers

   Vectors, similarity, and why keyword search is not enough.

   ~22 min account needed
8. 8
   RAG: answering from your own documents

   The architecture behind almost every serious AI product.

   ~25 min account needed
9. 9
   Making retrieval actually work

   Chunking, hybrid search, reranking and query rewriting.

   ~20 min account needed
10. 10
    Evals: knowing whether a change helped

    The discipline that separates engineering from prompt roulette.

    ~22 min account needed
11. 11
    Cost, latency and prompt caching

    Making it fast and affordable without making it worse.

    ~22 min account needed
12. 12
    Shipping: the things that break in production

    Secrets, retries, timeouts, abuse, monitoring and rollback.

    ~22 min account needed