Idea Explanation

Our idea is Gemma, a personalised AI tutor powered by Gemini that helps students solve problems through guided reasoning rather than instant answers. Instead of simply telling students the solution, Gemma supports them step by step with prompts, hints, and feedback that encourage them to think, revise, and learn actively.

What makes Gemma different is that it is built around how students actually learn. Users can type a question, upload a photo of handwritten work, or write directly on a whiteboard. Gemma then interprets the student’s working, responds conversationally, and adapts the level of support based on what the student appears to understand. The product also allows students to save and revisit previous work, making reflection and long-term progress part of the learning process rather than an afterthought.

In short, Gemma fixes the problem by turning AI from a shortcut into a support system. It makes tutoring more accessible while protecting the parts of learning that matter most: questioning, reasoning, revising, and building confidence

Implementation

Gemma is built as an AI-powered tutoring workspace with a clear separation between the frontend experience, the backend logic, and the AI services that power guidance and interaction.

On the frontend, students interact with a subject dashboard and tutoring workspace where they can type, upload, or write directly on a whiteboard. This interface captures the student’s input and sends it to the backend for processing. The frontend is responsible for presenting the learning workspace, showing prompts and hints, displaying saved work, and making the experience feel interactive and student-friendly.

Gemma works by receives requests from the frontend, processes the input, and decides which services to call. For example, when a student uploads an image of handwritten work or types a maths problem, the backend sends that content to the Gemini API, which analyses the input and generates guided hints, follow-up prompts, and adaptive tutoring responses. In the demo, we also use Cloud Text-to-Speech API so Gemma’s responses can be spoken aloud, making the tutoring experience more accessible and conversational.

A database would sit behind this system to store user accounts, learner profiles, subject workspaces, saved history, and revisitable attempts. This is important because Gemma is not meant to be a one-off chatbot; it is designed as a persistent learning environment where students can return to past work, track progress, and build understanding over time. In practice, the flow is:

1. Gemma captures a student’s typed, written, or uploaded input.

2. Gemma sends the relevant content to Gemini for interpretation and response generation.

3. Gemma optionally sends text output to Cloud TTS for spoken feedback.

4. The response is returned to the frontend, where the student sees or hears the next guided step.

5. User progress, work history, and saved sessions are stored in the database for future use.

This architecture allows Gemma to combine multimodal input, guided tutoring, and persistent learning records into one coherent system

Challenges

One of our biggest challenges was designing an AI tutor that felt genuinely educational rather than just another answer-generating chatbot. It is easy to build a system that gives responses, but much harder to build one that supports the student’s reasoning process without taking that process away from them. We had to think carefully about how to make Gemma feel helpful without making it dependent on answer dumping.

Another challenge was translating our concept into a realistic product flow. Because Gemma is interaction-heavy, we needed to think beyond a standard chat interface and create a workspace that supported typing, uploading, writing, speaking, and revisiting work. This required us to refine the user journey significantly so that the product felt like a real tutoring environment rather than a generic AI wrapper.

We also had to think about accessibility and multimodal learning from the beginning. This meant not only integrating Gemini for text and image understanding, but also considering voice-based interaction through Cloud TTS and designing for different learning preferences. We overcame these challenges by narrowing our demo scope, focusing on one clear tutoring flow, and making sure each technical choice supported our educational goal rather than distracting from it.

Accomplishments

Through this project, we learned how to turn a broad concern about AI and education into a clear product concept with a defined user need, technical structure, and future roadmap. We learned how to think more critically about the role of AI in learning, especially the difference between systems that optimise for fast answers and systems that support understanding.

We are particularly proud that we built a concept that is both technically credible and socially meaningful. Gemma is not just an AI demo; it is a product idea with a clear problem statement, a strong pedagogical angle, and a believable implementation pathway. We successfully combined guided reasoning, multimodal input, adaptive support, and accessibility into one coherent vision. We also developed a strong product identity, clear presentation narrative, and roadmap for how Gemma could evolve from MVP to a more personalised and inclusive learning platform.

The Future of Gemma

The next steps for Gemma would focus on deepening personalisation, expanding accessibility, and making the learning experience even more proactive.

First, we would continue building out personalisation features, including tutor customisation, voice and speed controls, reactive tutor avatars, progress analytics, and subject-specific workspaces. These features would help Gemma feel more engaging and more tailored to different learners.

Second, we would expand accessibility features, including speech-to-text, dyslexia- and ADHD-friendly modes, offline learning support, and wider subject coverage. This would make Gemma more inclusive for students with different learning needs and in different study environments.

Third, we would add smarter study tools, such as quiz mode for active recall, PDF import and annotation, editable worksheet conversion, and summarisation of learning documents. Finally, we see strong future potential in agentic AI, where Gemma could go beyond responding to prompts and begin proactively choosing the next best learning step, identifying misconceptions across sessions, and recommending targeted revision paths while still keeping the student in control of solving the problem.