Build Your First AI-Powered App
Build an AI writing assistant you fully understand.
About the Course
Most developers know AI is worth learning. What stops them isn't ambition, it's not knowing where to start. The terminology is everywhere, every tutorial reaches for a different wrapper library or framework, and "just use the API" doesn't mean much when you've never opened one. You end up with a borrowed demo and no real idea what happened, or which part you'd change to build your own.
This course strips that away. You'll build an AI writing assistant, a chat-style app where you type a request and it streams back drafts, edits, and feedback in real time. Two short sections up front build the mental model you need: how models work, and what tokens and context actually cost. From there, each building concept (system prompts, streaming, conversation state) arrives the moment the code needs it. Over eight sections you go from your first API call to a multi-turn app that streams token by token, and you understand every line.
The whole course is text. That's deliberate. You can copy a code block straight into your editor without scrubbing through a video, search a lesson for the parameter you half-remember, and come back to the streaming section months later when you wire it into your own project. It's the format you already reach for when you're stuck, turned into a course.
By the end you won't just have something to show. You'll have a real app you built yourself and the mental model to add AI to anything else: a CLI, a Slack bot, the side project sitting in a folder somewhere. This is the first course on the Mezie Labs AI Engineering Path, so it stays focused on the fundamentals (the API, prompts, parameters, streaming, and conversation state) and hands off the heavier topics, like tool use and retrieval, to the courses that build on it.
What You'll Build
You'll build an AI writing assistant: a single-page chat app where you type a request and it streams back drafts, edits, and feedback in real time. The backend is Node and Express talking to Claude through the official Anthropic SDK; the frontend is plain JavaScript with no framework and no build step. It handles multi-turn conversations, streams responses token by token, and fails gracefully when the API errors out.
What You'll Learn
- Explain how a language model generates text, and where it goes wrong
- Reason about tokens, context windows, and what an API call costs
- Make your first call to Claude with the official Anthropic SDK and read the response
- Shape the assistant's behavior with a system prompt and max_tokens
- Stream a reply token by token with
messages.stream()andres.write(), and render it live in the browser - Hold a multi-turn conversation by sending the full message history each turn
- Handle API failures without leaking error details to the client
- Apply prompt patterns like few-shot, and recognize when prompting isn't the right tool
Prerequisites
- Comfortable with JavaScript (ES6+), including async/await
- Able to run a Node project from the terminal (
npm install,npm run) - No prior AI or machine learning experience required
