NEXUS AI - Claude Code Tutorial

This Claude Code tutorial goes beyond the basics of installation, guiding you through building a real AI-powered document Q&A application from scratch. By the end, you'll have a working application deployed online, all within an afternoon.

Claude Code is Anthropic's AI coding assistant that operates at the project level. Unlike traditional tools that suggest lines within an editor, Claude Code reads your entire codebase, understands file relationships, and makes multi-file changes with full context. This means you describe what you want to build, and Claude Code handles writing the code, running commands, fixing errors, and iterating without switching between a chat window and your editor.

What sets Claude Code apart for AI app development is its ability to generate project boilerplate, write tests alongside generated code, catch its own mistakes by running the code and reading error output, and handle tedious tasks like CI config, requirements.txt setup, and test scaffolding. This allows you to focus on the core problem at hand.

To get started, you'll install Claude Code using `npm install -g @anthropic-ai/claude-code`, which requires Node.js 18 or higher. Once installed, authenticate Claude Code by opening a browser window during the first run, which will authenticate with your Anthropic account. After authentication, you'll enter an interactive session in your current directory.

The tutorial provides a step-by-step guide to build a document Q&A API using Claude Code. First, you'll scaffold a FastAPI project with a single `/health` endpoint, a `requirements.txt`, and a `.gitignore`. Claude Code will create the necessary files and confirm the actions.

Next, you'll build a real AI app using Claude Code. This involves creating a FastAPI application for document Q&A. The application will accept a `POST /upload` endpoint to store text files in memory and a `POST /ask` endpoint to answer questions using OpenAI's gpt-4o-mini model. The answers will be returned in JSON format with the answer text and a confidence score. Claude Code will generate the entire project structure, including the `main.py` file, which contains the endpoints, in-memory document store, and OpenAI call.

In the second step, you'll run the application locally using `uvicorn` and address any errors that arise. Claude Code will execute `uvicorn main:app --reload`, read the output, and automatically fix any import errors or missing packages. This iterative run, read error, and fix loop highlights Claude Code's efficiency and collaboration with the developer.

The tutorial then demonstrates how to improve the retrieval process for the `/ask` endpoint. Instead of sending the entire document to OpenAI on every request, Claude Code instructs you to split documents into 500-token chunks during upload, use cosine similarity with TF-IDF vectors to find the top 3 relevant chunks, and send only those chunks to OpenAI as context. This approach optimizes token usage and handles large files more efficiently.

Finally, you'll write tests for both endpoints using pytest. The tests will verify that a sample document is uploaded correctly, that a question about the uploaded document returns an accurate answer, and that querying a non-existent document returns a 404 error. Claude Code generates the test files, uses pytest-mock for mocking OpenAI calls, and updates `requirements.txt` with the necessary test dependencies. It runs the tests, reads the output, and iterates until all tests pass.

A unique feature of Claude Code is the `CLAUDE.md` file, which acts as a permanent briefing document for your AI collaborator. This file, located in the project root, documents the tech stack, coding conventions, test setup, and the functionality of the `/upload` and `/ask` endpoints. By maintaining this file, Claude Code can start each session with full context, eliminating the need to re-explain the project details in every new session.