Inspiration
In South Africa, Mathematics and Physical Sciences are gateway subjects, yet many learners drop them early because they are seen as too difficult or abstract. Large class sizes, limited lab access, and delayed feedback mean students often get stuck without help when it matters most. We were inspired to create a tool that gives learners immediate, visual, tutor-like support using the device they already have — a phone.
What it does
Funda AI Tutor is an AI-powered Lab Assistant built with Gemini 3. Students can take a photo of a lab setup, worksheet, graph, or math problem. Gemini analyzes the image and: Explains what the experiment or problem is about Guides learners step-by-step instead of giving full answers Checks student answers and explains mistakes Highlights common lab errors and safety considerations The app focuses on understanding, not memorization.
How we built it
Funda AI Tutor is built around Gemini 3’s multimodal API. We use Gemini 3 to: Interpret images of lab equipment, handwritten notes, graphs, and equations Perform deep, multi-step reasoning Generate structured, learner-friendly explanations Respond with low latency for a conversational experience The frontend allows users to capture or upload images and enter answers, while Gemini handles reasoning, evaluation, and explanation. Without Gemini’s image understanding and reasoning, the app would not function.
Challenges we ran into
Some of the main challenges included: Designing explanations that guide learners without revealing full solutions Handling varied image quality and handwritten inputs Balancing simplicity for beginners with depth for advanced learners We addressed these by refining prompts, structuring responses, and focusing on clear educational outcomes.
Accomplishments that we're proud of
Building a true multimodal education tool, not just a chatbot Demonstrating clear use of Gemini 3’s image understanding and reasoning Creating an experience that feels like a real tutor, not an answer engine Addressing a real educational challenge relevant to South Africa and beyond
What we learned
We learned that: Multimodal AI is most powerful when it guides thinking instead of giving answers Clear structure and prompt design greatly improve educational explanations AI can meaningfully support teachers and learners when designed responsibly This project deepened our understanding of Gemini 3’s strengths in reasoning and multimodal learning.
What's next for Funda AI Tutor
Next, we plan to: Add learner level adaptation (Beginner / Intermediate / Advanced) Expand lab safety detection and feedback Support more subjects and curriculum-aligned content Pilot the tool with real students and educators Our long-term vision is to make high-quality, AI-powered learning support accessible to every learner.
Log in or sign up for Devpost to join the conversation.