Introduction
I’ve been working with Semantic Kernel for the past year, and it’s completely changed the way I integrate AI into my applications. In this new blog series, I’ll walk through how to use Semantic Kernel in real-world projects, sharing the lessons I learned while building two apps:
Task Analytics, my own SaaS for Todoist task management and insights
Clobba, a Teams bot that enforces robust security rules in a multi-tenant environment
Before we get into specific demos, let’s talk about why I believe Semantic Kernel is one of the best tools for incorporating AI without losing control, reliability, or security.
Why Semantic Kernel?
1. Enhance Your App with Little Effort
Traditional AI integrations can be time-consuming. Even a seemingly simple feature, like controlling tasks with voice commands, can turn into a huge coding exercise—parsing user input, mapping it to database operations, and handling edge cases. With Semantic Kernel:
You teach it a few skills—in my case, marking tasks as complete or changing due dates.
You let users talk in plain language.
Semantic Kernel interprets the request and executes the right plugin.
That’s exactly how I approached Task Analytics. Instead of writing a large chunk of custom code, I leaned on Semantic Kernel to handle the complex language parsing. It became straightforward to let users say “Complete this task” or “Move my meeting to tomorrow,” and let the AI handle the rest.
2. Enforce Role-Based Security
Connecting an LLM directly to a database sounds tempting—libraries exist that can convert natural language into SQL queries automatically. But what happens if you have multiple customers, each with different permissions? Or different departments that shouldn’t see each other’s data? This is where Semantic Kernel’s fine-grained security controls shine.
Consider Clobba, a bot I built that interacts with data in a multi-tenant environment. If the AI just fed plain-text prompts into SQL, it would be easy for a manager in one department to accidentally (or maliciously) query sensitive information from another department. Semantic Kernel solves this by:
The result? A powerful bot that can answer a range of questions while respecting strict security boundaries—a must-have in enterprise settings.
Conclusion: Balancing AI Freedom and Control
Semantic Kernel provides exactly the right balance: you can let AI handle labor-intensive tasks—like language parsing, user-friendly interfaces, or data summaries—while still defining what it can and can’t touch. With role-based security and flexible skill creation, you maintain complete oversight, ensuring users don’t overstep their privileges or gain unintended access.
Looking Ahead: PeakIt #007 and Upcoming Posts
On April 3rd, I’ll be speaking at the PeakIt #007 conference in Brasov, sharing more about my journey with Semantic Kernel. The talk focuses on how developers can integrate AI in their apps in a responsible and useful way. If you’re curious about real-world AI adoption, I’d love to see you there.
Over the next few blog posts, I’ll show:
How to build secure, multi-tenant apps with Semantic Kernel using Clobba as an example, ensuring each user only has access to their data.
How to create user-friendly and useful interactions like I did in Task Analytics, letting users talk naturally to manage tasks without heavy custom code.
