You're launching "Future Bytes," your new tech podcast. Episode 1 is about the latest AI breakthroughs, but you've got 24 hours to:
- Research the topic
- Find credible sources
- Write a compelling script
- Make it sound natural
Plot twist: You don't have to do it alone. You're about to build your first AI assistant who can help with all of this. Let's call them Alex — your tireless research partner who never needs sleep.
Think of Small Language Models (SLMs) as your personal AI that lives on your computer. No cloud, no monthly fees, no sketchy data sharing.
Why SLMs are 🔥:
- 🏠 Runs on Your Machine: Laptop, desktop, even a beefy Raspberry Pi
- 💸 Zero Ongoing Costs: No API fees eating your lunch money
- 🔒 Privacy First: Your data never leaves your device
- ⚡ Lightning Fast: No internet lag, instant responses
- 🪦 Lightweight: 1B-10B parameters vs 100B+ for the big guys
Popular SLMs: Qwen 3, Phi-4, Gemma 3 (we're using Qwen for this workshop)
Ollama is like Steam for AI models. Download, run, and manage models with simple commands.
What makes it cool:
- One command to download and run any model
- Works on Mac, Windows, Linux
- Automatically uses your GPU if you have one
- Super memory-efficient
Microsoft Agent Framework is your playground for building AI agents that can:
- 💬 Chat and remember what you talked about
- 🛠️ Use custom tools (like searching the web or checking weather)
- 🧠 Think through complex problems step-by-step
- 🤝 Work with other agents as a team
- 🔌 Connect to different AI providers (OpenAI, Ollama, Azure)
The Building Blocks:
- Agents: Your AI assistants with specific jobs
- Tools: Special abilities you give them
- Memory: So they don't forget your conversation
- Reasoning: Teaching them to think, not just respond
The Quest: Build Alex, your podcast scriptwriter AI. Alex needs to generate dialogue between two hosts discussing tech topics.
What You'll Learn:
- How to wake up an AI agent (it's easier than waking up on Monday)
- Give it personality and instructions
- Make it generate actual podcast scripts
- Understand what it's saying back to you
Victory Condition: Alex creates a script for your "Future Bytes" pilot episode about AI! 🎯
The Quest: Alex is smart, but it doesn't know today's weather or what time it is. Let's fix that by giving it tools!
What You'll Learn:
- Create custom Python functions as "tools"
- Let Alex decide when to use which tool
- Watch it autonomously solve problems
- Combine multiple tools for complex tasks
Victory Condition: Ask "What's the weather in Tokyo?" and Alex figures it out on its own! ☁️
The Quest: Make Alex show its work. When solving problems, you want to see how it thinks, not just the answer.
What You'll Learn:
- Activate "reasoning mode" (it's like showing your work in math class)
- See Alex's step-by-step thought process
- Understand chain-of-thought prompting
- Debug when Alex gets confused
Victory Condition: Ask a tricky math problem and watch Alex think it through! 🧠
The Quest: Alex's knowledge has a cutoff date. Let's connect it to the web for real-time info!
What You'll Learn:
- Build a custom web search tool
- Integrate external APIs
- Handle network errors gracefully
- Get info beyond Alex's training data
Victory Condition: Ask about today's tech news and get fresh results! 📰
Required Gear:
- Python 3.10+ installed
- Ollama running (check with
ollama --version) - VS Code with Python extension
- At least 8GB RAM (16GB if you want smooth vibes)
Follow the notebooks in sequence for the full story:
- 00.BasicAgent-agent.ipynb — Meet Alex (your first agent)
- 01.BasicAgent-tools.ipynb — Power-up time!
- 02.BasicAgent-reasoning.ipynb — Teach Alex to think
- 03.BasicAgent-websearch.ipynb — Internet access unlocked!
After Act 1, you'll be able to:
- ✅ Run AI models on your own hardware (no cloud needed!)
- ✅ Build agents with custom personalities and skills
- ✅ Give agents tools to solve real-world problems
- ✅ Make agents show their reasoning process
- ✅ Connect agents to external data sources
- ✅ Debug when things go sideways
The Fix: Your computer's struggling. Try closing other apps, or switch to a smaller model. 8GB RAM is the bare minimum.
The Fix: Enable GPU acceleration in Ollama settings. Or reduce the context window size. Speed demon mode activated! 🏎️
The Fix: Double-check your function signatures. Alex needs proper type hints to understand what the tool does. Think of it like giving clear instructions.
- Agent Framework Docs — Official guides and examples
- Ollama Model Library — Browse all available models
- Qwen Model — Meet your AI's brain
- Code Examples — Steal ideas from here
You've got one agent. But what if you had a team of agents working together? In Act 2, you'll build your full podcast production crew:
- Researcher Agent: Finds the best sources
- Writer Agent: Crafts the perfect script
- Editor (You!): Approves or requests changes
Let's orchestrate some AI magic! → Act 2: Assemble Your Production Team
Stuck? Ask questions during the workshop. We're all learning together! 🙌