AIActBot, a 24-hour challenge by Sailpeak Lab.

At Sailpeak, we are more than consultants. We are doers, entrepreneurs at heart. That's why our team members get the opportunity to work on side projects when they are not working on client projects.

Recently, we threw a massive challenge to one of our Data Scientists, Léonard: Build a fully functional chatbot in just 24 hours.

And as we are big fans of AI and the AI Act became effective 3 months ago, we thought it would be nice to spare people the time to read a 459 page document to know all the ins and outs on the AI Act.

Léonard accepted the challenge to create the AIActBot in 1 day and didn't lose any time to get started.

Here’s how it went:

→ Stage 1: He considered building it from scratch. But he quickly realised the clock was ticking, and that this approach was a bit complex. So he iterated fast.

→ Stage 2: He leveraged existing “off-the-shelf” tools, custom solutions, and focused on accessibility. The goal was clarity, simplicity, and speed.

Curious how he did it step-by-step?

‍1) Retrieving the data
‍We began by sourcing the official document of the EU AI Act to ensure the chatbot would have a comprehensive knowledge base from trusted sources.

‍2) Storing the document
‍Using Google Cloud Storage, he securely stored the document, enabling scalable access and efficient data management for the project.

‍3) Building the AI Agent
‍Next, he used Vertex AI's Agent Builder to create an application. The stored document was integrated into the application's data store, laying the foundation for the chatbot’s AI capabilities.

‍4) Training the AI Model
‍Then, Léonard trained the advanced Gemini-1.0-Pro-001 model on the data store, allowing it to interpret and respond to complex legal queries regarding the EU AI Act.

‍5) Fine-tuning and customization
‍In Dialogflow CX, he fine-tuned the model to improve contextual understanding and set up guardrails. We also enhanced the user experience by customizing the interface with HTML and CSS.

‍6) Publishing and API Integration:‍
‍Finally, he published AIActBot and integrated it with the Agent API, enabling smooth real-time interactions for users seeking information on the EU AI Act.

By the next day, Leonard, our Data Scientist, got it done. The result was more than just a (very useful) chatbot. It was a powerful example of what relentless focus and ingenuity can create.

And of course, we didn't make it all for nothing. You can test it out and ask the AIActBot any question you'd have regarding the AI Act and it will give you a simply and straightforward answer within seconds. Feel free to test it out.