Inspiration

Modern living room media experiences can feel static and passive. When watching movies or TV shows, viewers often have instant questions about scene context, character motivations, or plot subtleties. FirePulse was built to turn television into an interactive experience on Android TV / Fire TV—giving viewers a seamless, real-time AI companion that answers scene-specific questions and generates instant trivia quizzes directly from their couch.

What it does

FirePulse connects a Kotlin-based Android TV frontend to a custom Node.js/Express backend integrated with Gemini API via OpenRouter:

Interactive Scene Q&A: Viewers click a button to query dynamic scene context, receiving quick and engaging AI-generated explanations tailored for TV viewing.
Dynamic Quiz Generation: Viewers can trigger interactive trivia quizzes based on the current scene context to test their knowledge and engage with content.
TV-Optimized Interface: Designed specifically for leanback viewing with clean TV navigation and fast response rendering.

How we built it

Frontend: Android TV / Fire TV application built in Kotlin, utilizing asynchronous network handling to communicate with local and deployed endpoints.
Backend Service: A fast Node.js & Express REST API server (server.js) deployed to handle request routing, CORS policies, and prompt structuring.
AI Integration: Integrated with Gemini via OpenRouter REST API endpoints for rapid, low-latency zero-shot reasoning and creative content generation.

Challenges we faced

API Key & Auth Compatibility: Resolved token type mismatches (401 ACCESS_TOKEN_TYPE_UNSUPPORTED) between modern SDK token handling and backend routing by pivoting to a robust REST architecture with OpenRouter.
TV UI Responsiveness: Ensuring network calls execute asynchronously so the TV interface remains smooth and never blocks viewer interactions or focus state.

What we learned

Building resilient backends by abstracting AI API layers through custom REST wrappers.
Fine-tuning prompt structures specifically for leanback smart TV user interfaces, keeping generated text concise and highly legible from a distance.

What's next for FirePulse

Real-time video frame context analysis using multimodal vision models.
Personalization based on viewer watching history and preferred quiz difficulty levels.