• 

  ui

Ariadne

A Voice-First AI companion for visually impaired

Inspiration

Visually impaired users face significant daily challenges, from navigating unfamiliar environments to identifying people and places. Existing technology often lacks accessibility-first design, making independence and confidence harder to achieve. Ariadne was inspired by the need for an intelligent, voice-first assistant that empowers users to navigate and interact with the world seamlessly.

What it does

Ariadne is an AI-powered mobile application that provides real-time assistance for visually impaired users. Key features include:

• Voice-first interaction: Users can speak naturally and receive spoken guidance.
• Audio & video recording: Capture surroundings and get contextual explanations.
• Bookmarks for people and places: Upload images to recognize individuals or locations later.
• Navigation assistance: Step-by-step voice guidance for safer and easier navigation.

How we built it

• Next.js based fullstack app: The core of Ariadne is a fullstack application built with Next.js.
• Bookmark system: Users can upload an image with a description, which is then uploaded to Gemini's files API with the display name set to the description. When a user makes a query, all bookmarks are loaded into Gemini with their descriptions. we can process large amount of bookmarks thanks to the large context window
• Audio and video processing: User audio and video recordings are sent to Gemini 3 for interpretation thanks to the multi model context.
• Voice Activity Detection (VAD): An algorithm monitors the user's voice levels. If the user is silent for 2 seconds, the recorded data is sent to Gemini, which interprets it and responds accordingly. this isnt handled by gemini as currently gemini3 doesnt support live api

Challenges we ran into

• Ensuring accurate recognition of people and places from uploaded images.
• Optimizing navigation guidance for real-world variability.
• Integrating Gemini Model 3 for real-time responsiveness.
• Designing an interface and experience that is fully accessible without visuals.

Accomplishments that we're proud of

• Successfully implemented a fully voice-driven AI assistant.
• Created a bookmarking system for personalized recognition.
• Developed step-by-step navigation optimized for visually impaired users.
• Leveraged Gemini Model 3 to provide intelligent and context-aware responses.

What we learned

• Accessibility needs to be a core design principle from the start.
• Real-time AI interactions require careful optimization to ensure responsiveness.
• Users benefit most from clear, concise, and contextually relevant feedback.
• Combining audio, visual, and AI processing can significantly enhance user independence.

What's next for Ariadne

• Integrate Google Maps API to provide more accurate navigation in public spaces.
• Connect the app to hardware devices to enhance real-world usability.
• Improve image recognition accuracy for bookmarks in different lighting and angles.
• Expand natural language capabilities for more complex queries.
• Conduct user testing with visually impaired individuals to refine UX and features.
• Explore offline functionality for environments with limited connectivity.

Built With

• gemini
• nextjs