Build a Production AI Agent in 6 Weeks

Beyond the hype: build a maintainable system around the AI

The agent concept is easy. Input, a decision, an action, in a loop. The hard part is everything around it: validation, boundaries, fallbacks, and the structure that makes it predictable instead of flaky. Reliability doesn't come from a smarter model. It comes from a clearer system.

This cohort is six weeks of building that system properly. You build Expense AI Agent, a complete application with function calling, a Telegram bot, a web dashboard, 95%+ test coverage, and a Docker deploy. You write the production code; we deeply review it for architecture, naming, and decoupling. Co-led with AI engineer Juanjo Expósito.

Who is this for?

Intermediate Python developers who want to go beyond API demos and LLM wrappers, and architect AI applications that are testable, deployable, and maintainable.

Time commitment: 6-10 hours per week.

What you'll build

A full-stack AI agent, from data layer to deployment, with three interfaces and production-grade engineering.

Agent architecture

Repository pattern, service layers, and Python Protocols for swappable LLM providers. Function calling and Pydantic structured outputs, not string parsing.

Three interfaces, one agent

A Typer + Rich CLI, a Telegram bot with human-in-the-loop confirmation, and a web layer (FastAPI REST API plus a Streamlit dashboard). One core, three ways in.

Production-grade testing

150+ tests, 95%+ coverage, Docker deployment, and CI/CD. You ship an app that's ready for users, not a notebook that's ready for a demo.

What you build, week by week

Week 1

Scaffolding

  • Repository pattern for data access
  • SQLModel entities & migrations
  • In-memory + database repositories
  • Test-driven from day one
Week 2

LLM Integration

  • Protocols for swappable providers
  • Pydantic structured outputs
  • OpenAI function calling / tools
  • LLM client abstraction layer
Week 3

Agent Tools & CLI

  • Prompt engineering for classification
  • Service layer orchestration
  • CLI with Typer + Rich
  • Classification pipeline end-to-end
Week 4

Telegram Bot

  • Input preprocessing & validation
  • Conversation state management
  • Human-in-the-loop confirmation
  • Mobile-first AI interface
Week 5

Web Interface

  • FastAPI with dependency injection
  • Pydantic request/response schemas
  • Streamlit dashboard + Plotly charts
  • REST API with OpenAPI docs
Week 6

Deploy & Ship

  • Docker multi-stage builds
  • Docker Compose orchestration
  • GitHub Actions CI/CD
  • Production-ready deployment

Tech stack: SQLModel, Pydantic, OpenAI function calling, Typer + Rich, FastAPI, Streamlit, Telegram bot API, Docker, GitHub Actions

Why this transfers

The patterns here are not AI-specific. They're how senior engineers build software: Protocols, repository pattern, service layers, dependency injection, clear boundaries, real tests. That's the difference between a demo that works once and a system you can run again and again with confidence. You leave with a deployed AI agent on your GitHub and the instinct to build the next one yourself.

How an AI expense agent is actually structured →

Real results

Jeff had decades in software but wanted a real map of AI. Six weeks later he'd built an agent with three interfaces and ~250 tests, and the mental model to go with it.

"It's different having done it than even reading about it. If somebody had written an explanation, a week later I'd have a vague idea, a month later I'd say I read an article about it. This, I think I understand."
— Jeff H., built an agentic app in 6 weeks

Why a coach, not just the model

An assistant will generate the agent for you. It won't tell you whether the architecture holds, expect your next PR, or give you the confidence to run this in front of users. Two coaches review every push, and you finish able to judge what the model produces instead of rubber-stamping it. That judgment is the job now, and it's learned from people who've built these systems, not from more output.

Join the cohort

Six weeks, two coaches, a portfolio-ready project. Capped small for high-touch, detailed PR review on every push.

€2,000 · one-time · 6 weeks · capped at 6 developers

Book a call to reserve your seat →

Full curriculum and enrollment: pythonagenticai.com. Prefer 1:1 on your own AI project? Start here →