Athena: Real-Time Multimodal AI Mentor

Athena AI uses Gemini 3 Flash & Pro for real-time multimodal mentorship. It leverages high-fidelity visual reasoning to analyze handwriting and code, providing instant context-aware student guidance.

Comment

Inspiration

Education is often one-size-fits-all, but learning is a deeply personal journey. We were inspired by Athena, the Greek goddess of wisdom, strategy, and handicraft. We wanted to build a tool that doesn't just "answer questions" but mentors the user—guiding them through problems, verifying their study, and providing real-time feedback on the world around them.

The goal was to create an AI that feels less like a search engine and more like a study partner who sits beside you. Whether you're solving a math problem on paper or need a quick explanation of a historical landmark, Athena is there to see what you see and help you understand it.

What it does

Athena AI Mentor is a multimodal AI companion built with Flutter. It uses the device's camera to "see" the user's environment in real-time.

  • Visual Analysis: Point your camera at a textbook, a diagram, or an object, and Athena analyzes it instantly.
  • Interactive Mentorship: Ask questions via voice or text, and receive context-aware answers.
  • Study Verification: It doesn't just take your word for it—Athena can generate quizzes based on your study material to ensure you've truly mastered the concept.

How we built it

We built the application using Flutter for a seamless, native experience on Android (and iOS). The core intelligence is powered by Google's Gemini 3.0 Pro, leveraging its advanced multimodal capabilities.

Key Technologies

  • Flutter & Dart: For the cross-platform UI and logic.
  • Google Generative AI SDK (google_generative_ai): To integrate Gemini 3.0 Pro. We implemented a robust fallback mechanism to Gemini 1.5 Pro to ensure high availability.
  • Camera Package: For capturing real-time image streams.
  • Riverpod: For reactive state management, ensuring the UI stays in sync with the AI's processing state.
  • Text-to-Speech (TTS): To give Athena a voice, making the interaction feel more natural and accessible.
  • Firebase & Supabase: Utilized for secure user authentication, cloud storage, and real-time database needs (managing user profiles, task logs, and chat history).

Architecture

We designed a clean, modular architecture. The GeminiService handles all interactions with the AI, managing API keys and model configurations. We optimized the requests by tuning generationConfig (e.g., limiting output tokens) to ensure low-latency responses, crucial for a "real-time" feeling.

Challenges we ran into

  • Real-time Latency: Sending high-quality images to an LLM can be slow. We had to balance image quality with response time, optimizing the image compression before sending it to the API.
  • Model Hallucinations: Early versions would sometimes misinterpret blurry text. We improved this by adding UI guides to help users frame their shots better and implementing a "retry" mechanism with a more robust model (Gemini 1.5 Pro) when the primary model struggled.
  • State Management: Coordinating the camera stream, AI processing, and UI updates (loading states, error handling) was complex. converting to Riverpod gave us a predictable unidirectional data flow that solved many "state spaghetti" issues.

Accomplishments that we're proud of

  • Seamless Fallback: Implementing the automatic switch between Gemini 3.0 and 1.5 ensure reliability during the demo.
  • Smooth UI: The "glassmorphism" design and fluid animations (using flutter_animate) make the app feel modern and premium.
  • Verifiable Learning: We're particularly proud of the "Quiz" feature, where the AI generates questions dynamically from the material it "sees", actively testing the user's knowledge.

What we learned

  • Multimodality is the future: Text-only interactions feel limiting once you experience the power of showing the AI what you're looking at.
  • Prompt Engineering is Key: Tweaking the system instructions for Gemini made a massive difference in whether it acted like a "bot" or a "mentor".
  • Flutter's Flexibility: Flutter allowed us to iterate rapidly on the UI while maintaining high performance for the camera feed.

What's next for Athena AI Mentor

  • Voice Mode: Fully hands-free voice interaction.
  • AR Overlay: Drawing explanations directly on the camera feed using Augmented Reality.
  • Classroom Mode: Allowing teachers to push assignments to student's "Athenas".

Built With

Share this project:

Updates