PROJECT CASE STUDY

01 / 03

Your Next Binge.

A Fire TV app for finding something you can actually watch in India.

BUILT WITH

  • React Native for Vega
  • TypeScript
  • TMDB API
  • Vector search

SYSTEM OVERVIEW

CATALOGTMDB · India availability
QUERYEmbedding server
RANKINGOn-device vector search
FALLBACKKeyword → local title
Simplified from the project’s documented architecture and behavior.

THE CHALLENGE

A recommendation only helps if it is available here.

Streaming catalogs vary by region, and a title that looks interesting is a dead end if it cannot be streamed, rented, or bought in India. Your Next Binge puts India availability at the center of discovery, then adds semantic search and a watchlist around that constraint.

01 / DISCOVERY

Filter by what can actually be watched.

Home discovery uses TMDB data scoped to India. Trending titles are requested through the discover endpoint because TMDB’s trending endpoint ignores the watch-region filter. Provider availability is grouped into Stream, Rent, and Buy so ad-supported options are not silently excluded.

The detail view carries those offers forward, and My List stores title snapshots so a saved list survives an app restart.

02 / SEARCH

A small FastAPI service encodes the query. The app compares that vector on-device against a bundled index of India-scoped titles, avoiding a network request for every ranking operation.

Search has explicit fallbacks: TMDB keyword results intersected with the bundled index, then local title matching if the network is unavailable. The interface reports which mode produced the results.

03 / TRADEOFFS

Ship only the data the index needs.

The bundled index contains TMDB IDs, titles, years, provider names, and numeric vectors. Descriptions and artwork are fetched at runtime rather than redistributed in the index. This keeps the local search corpus small and avoids bundling metadata that can be requested when needed.

Trailer playback was cut because the Vega Virtual Device image does not include the required WebView module. The detail view uses the backdrop instead.

OUTCOMES

What the build demonstrates

1,198

India-scoped titles in the bundled index

31–33 ms

Five semantic search gate queries on the VVD

Persistent

Watchlist restored after force-stop and relaunch

VALIDATION & SCOPE

What’s verified, and what remains.

  • Debug and Release builds installed and launched on the Vega Virtual Device.
  • The tested live home showed 14 trending titles and 14 titles available on selected services.
  • Physical Fire TV hardware was not tested. TMDB artwork was exercised but not pixel-inspected.