Build AI backends in Python,
from first run to production.

Track your progress.

Sign inor start with What an AI backend is

Full curriculum

OrientationPA3

Orientation

What an AI backend is

Setting up: Python 3.14, uv, and your editor

How to use this course

Module018

Python that reads well

Values, names, and the REPL

Strings, f-strings, and t-strings

Lists, dicts, sets, and tuples

Control flow and comprehensions

Functions, arguments, and closures

Modules, packages, and imports

Errors and exceptions

Files, JSON, and pathlib

Module028

Python for real programs

Classes and dataclasses

Type hints that pay off

Generators and iterators

Decorators and context managers

The standard library you will actually use

Dependencies and environments with uv

Testing with pytest

Project layout and pyproject.toml

Module035

Async Python

Why async: waiting is the bottleneck

async, await, tasks, and gather

Timeouts, cancellation, and semaphores

httpx: async HTTP clients

Threads, processes, and when async is wrong

Module045

Pydantic

Models and validation

Field constraints and custom validators

Serialisation and JSON Schema

Settings and configuration

TypeAdapter and discriminated unions

Module059

FastAPI fundamentals

Your first FastAPI app

Path, query, and body parameters

Response models and status codes

Dependency injection

Routers and project structure

Middleware, CORS, and lifespan

Errors and exception handlers

Testing FastAPI

Auth: API keys, JWT, and OAuth2

Module065

Databases from Python

SQLAlchemy 2.0 async basics

Models, sessions, and Alembic migrations

Repositories and transactions in FastAPI

Redis for caching, rate limits, and queues

Background work: BackgroundTasks, ARQ, and Celery

Module075

Real-time: streaming and WebSockets

Streaming responses and JSON lines

Server-sent events for token streams

WebSockets in FastAPI

Connection managers and broadcasting

Backpressure, heartbeats, and reconnects

Module089

LLM APIs from Python

How an LLM call works: tokens, context, and cost

The OpenAI Responses API

The Anthropic Messages API

Streaming tokens to the browser

Structured outputs with Pydantic

Tool calling by hand: the loop

Prompts as code: templates, versions, and tests

Embeddings and a small RAG endpoint

Retries, timeouts, and rate limits for model calls

Module095

Agent fundamentals

What an agent is: loop, tools, state

Designing tools an LLM can use

Memory: short-term, long-term, and sessions

Guardrails and human approval

Evaluating agents

Module105

OpenAI Agents SDK

Agent, Runner, and the first run

Function tools and hosted tools

Handoffs and multi-agent systems

Guardrails, sessions, and tracing

Streaming an agent into FastAPI

Module115

LangChain and LangGraph

LangChain 1.0: models, tools, and create_agent

LangGraph: state, nodes, and edges

Persistence and human-in-the-loop

Streaming and structured responses

Choosing an agent framework

Module123

Model Context Protocol

MCP: servers, clients, and tools

Build an MCP server in Python

Connect agents to MCP servers

Module135

Observability with Langfuse

Why tracing an LLM app is different

Langfuse setup and the OpenAI drop-in

Tracing FastAPI and agents with @observe

Sessions, users, scores, and evals

Prompt management and self-hosting Langfuse

Module146

Shipping the AI backend

Configuration, secrets, and environments

Docker for Python services

Deploying: a VM, a container platform, or serverless

Cost control: caching, model routing, and budgets

Security: prompt injection, PII, and tool permissions

Capstone: a support-agent backend end to end

The free lessons are complete lessons, not teasers. Pro unlocks the rest of the path, including the agent, MCP, observability and deployment modules.Read a free lesson

124 online