Part 2: API Integration

Build real AI applications by mastering streaming chat, structured data extraction, and multimodal vision APIs.

What You'll Learn

This section moves from theory to practice. You'll integrate with OpenAI, Claude, and Gemini APIs to build production-ready features: streaming chat interfaces that feel instant, parsers that extract structured data from any document, and vision tools that analyze images. By the end, you'll have working code for the most common AI integration patterns.

Lesson 5: Text Generation & Streaming UIs

Build your first real AI integration: a streaming chat interface that works across multiple providers. You'll learn the anatomy of chat completion requests, implement Server-Sent Events for token-by-token streaming, and build a FastAPI backend that proxies LLM responses to your frontend. We'll cover provider abstraction so your code doesn't lock into one vendor, error handling with retry logic, and the patterns that make streaming UIs feel responsive. By the end, you'll have a production-ready chat API that streams responses in real-time.

Lesson 6: Structured Data Extraction

Transform messy documents into typed objects your code can trust. You'll use MarkItDown to convert PDFs, Word docs, spreadsheets, and presentations into clean Markdown—the format LLMs understand best. Then you'll leverage JSON mode and Pydantic schemas to extract structured entities with validation and retry logic. We'll build a document parser that extracts invoices, resumes, and contracts into typed Python objects. This lesson bridges the gap between unstructured human documents and structured machine data.

Lesson 7: Vision & Multimodal Inputs

Teach your LLM to see. Modern models like GPT-4o, Claude, and Gemini process images alongside text—enabling screenshot analysis, document OCR, diagram understanding, and visual Q&A. You'll learn image encoding strategies (Base64 vs URLs), cost optimization through sizing and tiling, and provider-specific vision APIs. We'll build practical tools: screenshot-to-code generators, receipt scanners, and UI analyzers. By the end, you'll know when vision adds value and how to implement it without breaking your token budget.

Why This Matters

These three lessons form the core of modern AI product development. Streaming makes your UIs feel fast. Structured extraction turns documents into data pipelines. Vision unlocks entirely new product categories. Master these patterns and you can build 80% of AI features shipped today.

Complete these lessons before moving to production topics in Part 3.