Welcome to my new book about building effective LLM-based applications with Semantic Kernel. This book is available on Leanpub! This repository is where I build the book.
Get a copy now and start reading! Read it on Leanpub: Building effective LLM-based applications with Semantic Kernel
I'm writing this book as you read this, and I'm at 43% completion.
| Chapter | Status | Samples |
|---|---|---|
| 1. Understanding Large Language Models | Complete | N.A. |
| 2. Essential LLMOps knowledge | Complete | N.A. |
| 3. Getting started with Semantic Kernel | Complete | C#, Python |
| 4. The art and nonsense of prompt engineering | Complete | C#, Python |
| 5. Testing and monitoring prompts | Complete | C# |
| 6. Enhancing LLMs with functions | Complete | C# |
| 7. Retrieval Augmented Generation (RAG) | Not started | |
| 8. Working with structured output | Not started | |
| 9. Prompt chaining workflows | Not started | |
| 10. Intelligent request routing workflows | Not started | |
| 11. LLM orchestration workflows | Not started | |
| 12. Artist and critic workflows | Not started | |
| 13. Building basic agents | Not started | |
| 14. Multi-agent teams | Not started |
This book features useful examples that demonstrate how to implement various patterns with Semantic Kernel. To use the examples you'll need:
- .NET 9
- Visual Studio Code
- Access to the OpenAI API or Azure OpenAI
- Azure CLI
You can open the solutions in the samples directory to browse through them.
Follow the instructions provided with each example to run the example.
Important: I haven't started writing the python and Java samples yet. I'll update the samples directory in between writing the chapters of the book.
I am writing this in my spare time to help people gain practical knowledge about working with large language models. The book is free, but I take donations as a token of gratitude.
You can donate by purchasing the book online through Leanpub. You can also sponsor me through the sponsorship button on this repository.
Since you're here anyway, let me share my workflow for this book so you'll get to know a little more about how I work and what you can expect from me.
My workflow has the following phases:
- Research
- Writing
- Review and initial editing
- Formatting and layout
- Publishing
I publish after each chapter is complete and undergone initial editing. This way, you get the content as fast as possible. After I've completed the full book, I'll go through the entire book again to ensure everything is consistent and correct.
Hey, didn't you use AI to write your book?
Yes, I did! I used Claude.ai to help me generate text for the book. But I stopped. I found that it gets in the way of critical thinking and creativity. It also throws so much mediocre content at me it's hard to manage it. I prefer to write the content myself now that I have the time for it. I love writing, and AI wasn't doing me a favor in that regard.
The book contains my experience with building LLM-based applications using Semantic Kernel. I try to pull from experience as much as possible because it gives the best results for me.
However, before I write, I do my research to verify that my ideas are going to work and build a sample to demonstrate the principles. Working code is still the best way to demonstrate things. The book is the narrative to the example.
I write an initial set of notes in Markdown and a rough outline that I refine as I go.
After the initial outline, I'll start by adding detail to each of the sections in the outline. I still use a note writing style as I want to make sure that each section makes sense to you as the reader of my work. Often, I need to rework the outline as I write the notes because I find that the initial outline doesn't work out as I thought.
Once I have the notes for the full chapter, I'll go back and expand the notes into the full text. It's at this time I want to make sure that there are pointers and bridges to help you navigate the text. I also want to make sure that the text is engaging and that it's easy to read.
Once everything is in the chapter, I'll put aside the text for a day and let it rot, ripen, or marinate if you will. This helps me to get a fresh perspective on the text. I'll then go through the text and make sure that everything makes sense and that the text still conveys what I wanted to tell you.
I perform a final spell check using Grammarly, an online tool that helps me catch mistakes in grammar, spelling, and style.
At this point I use the Leanpub preview generator to preview the final look of my book. I find that I often need to insert additional line breaks and correct figures to make sure that everything looks good. I also need to make sure that the layout of the code samples isn't broken.
When I think the chapter is ready, I'll publish a new version in Leanpub. I'll also update the README.md file to reflect the current state of the book.
