Smart Learning, Smarter Kids
AI That Helps You Think
We recently had the opportunity to participate in “The Agentic Era Hackathon”, an exciting experience focused on Google’s most recent AI product. This massive EMEA-wide event, orchestrated by Google Cloud’s EMEA Partner Engineer Practice for AI, brought together developers from across EMEA to tackle the challenge of building AI agents on Google Cloud. Armed with Google’s Agent Starter Pack, we set out to fast-track generative AI solutions to production. Here’s a look at what the day involved and what our team created…
The Challenge/use case
Today, many AI applications take over human effort by writing entire essays, blog posts, articles, and even complete academic assignments. These are tasks that traditionally relied on human reasoning and creativity. While this technology is impressive, it raises important concerns about the long-term impact on students’ ability to think critically and solve problems independently.
We believe AI should not replace thinking, but enable it. It should be a tool that strengthens understanding and deepens learning, not one that encourages passivity. Our vision is that students must remain the ones doing the thinking, while AI helps guide them in the right direction.
This mindset led us to focus on education. In Belgium, schools face a growing shortage of teachers. This puts pressure on the education system and limits the ability to give students the personal guidance they often need. Teachers have limited time and resources, and creating and correcting individual exercises is both time-consuming and repetitive. Meanwhile, students struggle to get targeted practice based on their specific needs.
Our challenge was to find a way to use AI so that it supports both students and teachers. We wanted to help students improve through active engagement and targeted feedback, while also reducing the workload for teachers by automating the generation, evaluation and adaptation of exercises, without ever replacing the human aspect of learning!
Our Solution
The solution we came up with is an AI-powered tutoring agent designed to help students actively practice and deepen their understanding of school subjects. Unlike tools that simply provide answers, this agent guides students through the learning process, encouraging them to think for themselves.
To ensure that the exercises are relevant and aligned with official learning goals, all study materials and curriculum objectives are stored in Google Cloud Storage. This setup allows the agent to retrieve only the most appropriate content for each student’s selected topic.
The process begins when a student chooses a subject and topic they want to work on. Based on this input, the agent pulls the relevant materials from storage and uses Gemini to generate tailored exercises. These questions are not generic or random — they are directly linked to curriculum objectives and adapt to the student’s level of understanding.
Once an exercise is generated, the student provides their response directly within the platform. The agent then evaluates the answer in real-time, offering constructive feedback, hints, and a score. This immediate feedback loop helps students quickly recognize and learn from their mistakes.
In the background, all performance data is automatically logged in BigQuery. This enables teachers and students to track progress over time, identify strengths and weaknesses, and gain valuable insights into learning patterns using tools like Looker Studio.
For students, this means access to personalized practice, real-time guidance, and the opportunity to close knowledge gaps independently. For teachers, it significantly reduces the time spent creating and grading exercises, while still offering full control over curriculum content and access to actionable data to support student learning more effectively.
System architecture
Our solution is built as a modular cloud-native application. The frontend (Streamlit) allows students to interact with the agent, selecting a subject and topic to practice. This request is processed by a RAG pipeline that retrieves relevant materials from Google Cloud Storage and uses Gemini for question generation and evaluation. All user interactions and performance data are stored in BigQuery, enabling learning analytics via Looker Studio.
From a non-functional perspective, the system includes observability through structured logging, built-in monitoring, and robust error handling. Data privacy is maintained using Google Cloud’s security and access control mechanisms.
What makes our solution special is the combination of curriculum-based content, personalised exercises, and real-time feedback. All of this in a setup that supports both students and teachers in a simple and efficient way.
✨ Important features and main elements of your solution
A cornerstone of our solution is the powerful analytics and reporting dashboard we���ve built. By seamlessly integrating BigQuery directly into our agent and leveraging Looker Studio for insightful visualizations, we provide educators with real-time visibility into student performance. This enables data-driven decisions that can truly transform learning outcomes. Coupled with Google’s latest breakthroughs in agentic AI, including the RAG Engine for enhanced information retrieval, the Google ADK for streamlined agent development, and the Agent Engine for efficient deployment and management, we’ve created a state-of-the-art, scalable learning platform.
🚀 Implementation and Agent Starter Pack
We found the Agent Starter Pack to be an incredible resource for accelerating our agent development. Having a clear understanding of our desired agent functionality and architectural design was essential. With this roadmap, we were able to strategically utilize the Starter Pack’s templates and quickly establish a foundation. From there, customizing and personalizing the agent became a straightforward process.
Taking Your Agent to Production and Next Steps
With the Smart Learning agent fully developed, rigorously tested, and validated, we are set for deployment. Our robust CI/CD pipeline is designed to manage this transition seamlessly. By simply pushing our code to the main branch, the automated deployment process is initiated, swiftly and reliably moving our Smart Learning agent into the production environment. This crucial step means we can now make this empowering tutoring tool immediately available to students and teachers. Most importantly, with the Smart Learning agent now deployed, we are excited to see students embark on a journey of personalized learning, empowered to think critically and achieve their full potential.
Conclusion
Our experience at the ‘Agentic Era Hackathon’ was incredibly fruitful. We successfully built a Smart Learning agent designed to empower student thinking, not replace it. This agent delivers personalized practice, adaptive exercises, real-time feedback, and valuable data insights for educators. We leveraged Google’s Agent Starter Pack and cutting-edge AI technologies to bring this vision to life.
We’re excited about the potential of AI to transform education and plan to continue refining this project.
Explore our code and join us! Find our repository here: https://github.com/jossehuybrechts/smart-learning. We welcome your feedback and contributions.
Let’s build the future of AI-powered learning together!
Authors:
, and
