Scene Sync

SceneSync — Catch Up Together. Never Spoil the Moment.

Inspiration

SceneSync was inspired by a familiar problem: friends and family often watch the same series at different speeds. Someone might be several episodes ahead, while another person is still catching up. Discussing the show becomes difficult because even a casual conversation can reveal a major plot twist.

We wanted to make watching together possible even when everyone isn't on the same episode. Our goal was to build more than a watch-party app: a spoiler-aware companion that understands each viewer's progress and helps everyone stay part of the conversation without ruining the suspense.

What It Does

SceneSync combines a native Amazon Fire TV application, a Chrome-compatible browser extension, and a smartphone companion into one connected experience. It is designed to work alongside streaming platforms such as Netflix, Prime Video, and JioHotstar.

The system tracks viewing progress, synchronizes watch-party members in real time, and provides AI-powered recaps and targeted Q&A. Its key feature is a spoiler-safety engine that establishes the earliest viewing point among active participants and restricts AI responses to information available up to that point.

Instead of relying solely on a prompt telling the AI not to spoil anything, SceneSync filters future-episode information before it reaches the model and validates generated answers against authorized evidence.

How We Built It

We designed SceneSync as a multi-platform system with a shared backend:

Fire TV application: A native Android TV experience optimized for the television screen, with large text, high-contrast focus indicators, and remote-control navigation.
Browser extension: A Chrome Manifest V3 extension with streaming-provider adapters for detecting show and episode context on supported streaming pages.
Smartphone companion: A mobile-friendly interface for joining watch parties, selecting viewing progress, controlling linked extensions, and asking questions.
Backend: A Fastify REST API, Socket.IO for real-time communication, and PostgreSQL for persistent data and transactional operations.
AI and spoiler protection: A provider-agnostic AI service with evidence filtering, citation validation, safe fallbacks, and cache isolation based on the authorized viewing boundary.

We also developed automated tests for viewing-boundary calculations, cross-device synchronization, streaming-provider detection, and adversarial attempts to extract future plot information.

Challenges We Faced

The biggest challenge was making spoiler protection a system-level guarantee rather than simply an instruction to an AI model. Even a well-written prompt can produce an answer containing information from later episodes. We addressed this by filtering the evidence before generation and validating the claims used in the final response.

Another challenge was keeping multiple devices synchronized. A viewer can join late, leave temporarily, or move backward to an earlier episode. SceneSync must recalculate the shared viewing boundary whenever relevant progress changes, ensuring that one viewer moving ahead does not expose spoilers to everyone else.

Supporting different streaming providers introduced another layer of complexity because their page structures and metadata differ. We designed separate provider adapters and a fallback mechanism to handle missing or ambiguous episode information.

Finally, building for Fire TV required a different design mindset from building for a phone or desktop. Navigation must work naturally with a remote control, and the interface must remain readable and accessible from across a room.

What We Learned

Building SceneSync helped us understand how to design a product across multiple platforms while maintaining a consistent source of truth. We explored real-time event-driven architecture, database transactions, browser-extension APIs, Android TV development, and AI integration.

Most importantly, we learned that responsible AI requires more than good prompts. Reliable output depends on controlling the information available to the model, validating its claims, and failing safely when an answer cannot be verified.

What's Next

We want to strengthen real-world streaming-provider compatibility, improve automatic episode detection, refine the Fire TV experience, and expand the catalog of verified episode information. We also aim to validate spoiler protection and synchronization across a wider range of devices and real watch-party scenarios.

SceneSync is built around one simple idea: everyone should be able to enjoy the same story together, regardless of how quickly they watch it.