Builder foundations for AI engineers
Every AI system is software first, so before models you need the skills that every build depends on: Python, calling APIs reliably, async code, data formats, Git and GitHub, the terminal, Docker, a first cloud deployment, keeping secrets safe, and validating data with Pydantic. These are the skills that decide whether your project works only on your laptop or works for other people. You don't need to master them before your first AI build, but you will use every one of them in the first month.
Why foundations come before models
When people start AI engineering, they want to jump straight to LLMs, RAG and agents, and I understand why, because that is the exciting part. But when you look at a real AI system, the model call is a small part of the code. Around it there is an API that receives requests, code that reads and cleans data, a database or vector store, error handling, logging, a container, a deployment, environment variables for the keys and a repository where the team works together.
So when a build fails, it usually doesn't fail inside the model. It fails because a library version is different on the server, an API key was missing, a request timed out, the output wasn't valid JSON, or someone pushed a .env file to GitHub. These pages are about those failures, and about the habits that prevent them.
Where each skill shows up in a build
The easiest way to see why these skills matter is to follow a project through the phases it goes through on its way to production. In our sessions, I describe six of them.
- Development and local testing: you write the code in an editor like VS Code, build the API, and test it on your machine.
- Containerization: you package the application with Docker so it runs the same way everywhere.
- Cloud setup: you choose a provider, like AWS, Google Cloud, Azure or a simpler platform, and set up what the application needs.
- Deployment: you push the container to that cloud so other people can use it.
- Frontend (optional): you add an interface if the users need one, from plain HTML to a Streamlit app.
- Monitoring, security and maintenance: you watch logs, costs and errors, protect the system and keep it running.
Each page in this section belongs to one or more of these phases.
| Page | What it covers | Phase |
|---|---|---|
| Python for AI engineers | The Python you will actually use, virtual environments, project structure | 1 |
| Working with LLM APIs | How APIs work, retries, rate limits, timeouts, streaming, logging | 1, 6 |
| Async Python for LLM apps | Running many slow calls at the same time without blocking | 1, 4 |
| Data formats for AI work | JSON, JSONL, CSV, Parquet, saving models and embeddings | 1 |
| Git and GitHub for AI projects | Version control, what to commit and ignore, collaboration, your profile | 1 to 6 |
| Linux and the terminal | The commands you use every day, servers, logs and long jobs | 1 to 6 |
| Docker for AI apps | Containers, Dockerfiles, Docker Compose | 2 |
| Cloud basics: your first deploy | Choosing where to deploy, and the steps of a first deployment | 3, 4 |
| Environment variables and secrets | Keeping API keys out of code and out of Git | 1, 4, 6 |
| Pydantic and structured data | Validating model outputs, tool arguments and API payloads | 1, 6 |
How to use this section
If you are new to programming, read the Python page first and spend real time on it, because everything else assumes you can write and run a small program.
If you already work as a developer, skim the pages and go straight to the parts that are specific to AI work: the retries and token logging on the API page, async for concurrent LLM calls, JSONL for datasets and logs, and Pydantic for validating model outputs.
Either way, don't try to finish this section before you build anything. Open each page when your build needs it, which in the first 30 days will be quite soon for all of them.
What these skills are not
These pages are not a full course on Python, Git, Linux or Docker. Each of those has whole books written about it. They cover the part of each skill that an AI engineer uses every week, explain why it matters for AI projects in particular, and link to the official documentation when you want to go deeper.
From the floor
In the 75 day AI engineering challenge, the first 7 days were only foundations: Python, the terminal, Git and GitHub, APIs, saving models, deployment with Flask, and Docker. I told everyone those 7 days were the most important ones, because if your fundamentals are set up right, every other day becomes easier.
When we restarted the AI Engineer HQ curriculum, I moved APIs and Docker from the project sessions into the fundamentals. Before that, builders met them for the first time in the middle of a project, while also trying to understand the AI part, and I think understanding the fundamentals of both on their own first is the better way to learn them.