G for future

Building Tomorrow, the Google Way

Comment

InspirationThe "2 Sigma" Problem: We were inspired by Benjamin Bloom’s research showing that 1-on-1 tutoring creates students who perform 98% better than those in standard classrooms.The Accessibility Gap: Quality private tutoring is a luxury that few can afford, leaving millions of students behind.Democratizing Intelligence: We wanted to ask: What if the world's best tutor lived in everyone's pocket?Google's Mission: We aligned our vision with organizing the world's information—specifically, making complex academic concepts universally accessible and intuitive.What it doesVisual Learning: Students can snap a photo of a textbook diagram, a handwritten equation, or a complex physics problem.Multimodal Analysis: G for Future uses Gemini 3 Flash to "see" and understand the context of the image immediately.Socratic Teaching: Instead of just giving the answer, the AI acts as a mentor. It breaks down the problem, identifies the core concept, and asks guiding questions to help the student learn.Mathematical Precision: The app renders complex math in beautiful LaTeX, transforming messy input into clean equations like:$$x = \frac{-b \pm \sqrt{b^2 - 4ac}}{2a}$$"Future Fact" Engine: Every explanation ends with a real-world application (e.g., linking calculus to machine learning algorithms) to inspire the student about the future.How we built itAI Core: We leveraged the Gemini 3 Flash API for its ultra-low latency and superior multimodal (vision + text) capabilities.System Architecture: We used Firebase Functions to create a secure, serverless backend that handles API requests without exposing keys.Frontend: We built a responsive, mobile-first web app using React and Vite to ensure a snappy user experience.Data Persistence: Firebase Firestore allows users to save their "Aha!" moments and revisit past sessions for exam revision.Prompt Engineering: We utilized "Chain-of-Thought" prompting and LearnLM principles to ensure the AI behaves like a teacher, not a search engine.Challenges we ran intoThe "Calculator" Trap: Initially, the model wanted to solve problems instantly. We had to rigorously tune the System Instructions to force it to "teach" rather than "solve."Handwriting Recognition: Parsing messy student handwriting was difficult. We overcame this by using Gemini’s vision capabilities to first transcribe the problem before solving it.LaTeX Formatting: ensuring the AI consistently outputted valid LaTeX code inside Markdown blocks required specific, iterative prompt constraints.Balancing Latency: We struggled to keep the response time under 3 seconds while maintaining high-quality explanations, eventually solving this by switching to the Flash model.Accomplishments that we're proud ofSeamless Multimodality: We successfully built a flow where a user can upload a photo and get an instant, context-aware explanation.Educational Guardrails: We created a safe environment where the AI refuses to just "do the homework" and instead fosters genuine learning.Speed to Deploy: We went from a concept on a napkin to a live, hosted application on Firebase in under 24 hours.Visual Polish: Integrating Material Design 3 components gave the app a professional, "Google-native" look and feel.What we learnedPrompting is Programming: We learned that natural language instructions are just as critical as the Python/JS code wrapping them.Visual Context Matters: Text-only inputs are insufficient for STEM education; the ability for the AI to "see" the diagram is the game-changer.Simplicity Scales: By stripping away complex logins and unnecessary features, we focused entirely on the core value: Understanding.What's next for G for FutureGemini Live Integration: Enabling real-time voice conversations so students can "talk through" a problem while walking to school.AR Overlay: Using Augmented Reality to project solution steps directly onto the physical textbook page.Vernacular Support: Expanding the model to detect and respond in regional languages (Hindi, Spanish, Telugu) to truly serve the "Next Billion Users."Teacher Mode: Creating a dashboard for educators to see aggregated data on where their students are struggling.

Built With

Share this project:

Updates