Building Generative AI Applications from First Principles

A workshop repository for learning to build AI-powered applications through hands-on experience.

Description:

This workshop teaches data and software folk how to build and iterate on generative AI applications using first-principles thinking.

What You'll Learn:

• How to integrate AI models and APIs into practical applications
• Techniques to manage non-determinism and optimize outputs through prompt engineering
• How to monitor, log, and evaluate AI systems to ensure reliability
• The importance of structured outputs and function calling in AI models
• The software engineering side of building AI systems, including iterative development and debugging
• Multi-LLM workflows and agentic patterns

Workshop Structure

The workshop is divided into 4 hands-on notebooks, guiding you from basic queries to complex agentic workflows:

• Notebook 1: Abstractions vs. First Principles

  • Build a simple RAG app using LlamaIndex.
  • Peel back the layers to rebuild it in Vanilla Python.
  • Add a frontend (Gradio) and observability (SQLite/Datasette).
  • Note: We'll be running standalone Python applications from the apps/ directory for this section.
  • Key Takeaway: Understanding what actually happens inside an LLM app by removing the abstractions.

• Notebook 2: Prompt Engineering & Evaluation

  • Build a two-stage pipeline: Extract structured data from LinkedIn profiles -> Generate personalized emails.
  • Implement an LLM-as-a-Judge to automatically evaluate email quality.
  • Key Takeaway: How to control output structure and automate quality testing.

• Notebook 3: Function Calling (Tool Use)

  • Teach LLMs to use external tools (Weather API, Tavily Search).
  • Understand the "Define -> Request -> Execute -> Respond" loop.
  • Key Takeaway: Connecting LLMs to the real world and enriching data.

• Notebook 4: Building Agents

  • Combine everything into an autonomous Agent.
  • Build a ReAct agent that can reason, plan, and use tools to solve multi-step problems.
  • Key Takeaway: Moving from linear pipelines to dynamic, autonomous systems.

Note: There are accompanying slides that will be shared during the workshop.

Prerequisites:

• Basic Python programming knowledge
• Familiarity with REST APIs
• Experience with Jupyter Notebooks (preferred but not required)
• No prior AI/ML experience required

Setup Instructions

Option 1: GitHub Codespaces (Recommended)

1. Create a Codespace

   • Click Code → Create Codespace on main

2. Activate the Environment

   source .venv/bin/activate

3. Set Up API Keys

   Create a .env file from the example:

   cp .env.example .env

   Then open .env and add your API keys:

   export OPENAI_API_KEY="your-key-here"
   export ANTHROPIC_API_KEY="your-key-here"
   export GOOGLE_API_KEY="your-key-here"
   export TAVILY_API_KEY="your-key-here"

Option 2: Local Setup

1. Create Virtual Environment

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

2. Install Dependencies

   # Recommended: Install with UV (faster)
   pip install uv
   uv pip install -r requirements.txt
   
   # Or use regular pip
   pip install -r requirements.txt

3. Set Up API Keys

   Create a .env file from the example:

   cp .env.example .env

   Then open .env and add your API keys:

   export OPENAI_API_KEY="your-key-here"
   export ANTHROPIC_API_KEY="your-key-here"
   export GOOGLE_API_KEY="your-key-here"
   export TAVILY_API_KEY="your-key-here"

---

📚 Continue Learning

Want to go deeper? Check out our Building AI Applications course. It’s a live cohort with hands-on exercises and office hours.

Get 35% off with this link for workshop participants.