VER-DÉ — AI Cinema for Fire TV

Tell your TV how you feel. VER-DÉ, a Fire TV app, finds the independent film for the moment, explains why, and remembers what your household loves.

Comment

Inspiration

Some nights you don't know what you want to watch. You only know what you feel.

I'm an independent filmmaker and farmer from Nabua, Camarines Sur, Philippines. I make films under the name VER-DÉ. My films are real, finished, and free or affordable to watch — but almost nobody finds them. Streaming apps are built around browsing posters and genres, and independent films get buried under catalogs of thousands.

So I asked: what if the TV asked how you feel, instead of what genre you want?

What it does

VER-DÉ for Fire TV is a remote-first, AI-powered way to discover independent films on the television.

  1. Who's watching? — Just me, With family, or With someone.
  2. What does the moment call for? — pick a mood with the D-pad: Something hopeful, About the land, Being judged, Surprise me… No typing, no menus.
  3. VER-DÉ understands — the request is interpreted into concepts, moods, runtime and language, and shown back as tags so the viewer sees what VER-DÉ heard.
  4. Three films, three reasons — not three hundred. Each card explains why it fits, grounded only in real catalog facts.
  5. Refine — Another one, Something shorter, Why this one?, Change the mood.
  6. Watch — free films play full screen. Founding Access films show a QR code so the viewer can support the filmmaker directly from their phone.
  7. The next evening, it remembers — each household's choices are remembered on the TV. VER-DÉ greets the family with what they watched last, offers "Something like it," and re-ranks recommendations toward what that household enjoys while favoring films they haven't seen.

How I built it

  • Fire OS app (Java, Android) — a full-screen WebView host with a home-screen banner, leanback launcher support, autoplay for playback, an offline screen, and a Back-key bridge that steps back through VER-DÉ's own screens before exiting.
  • 10-foot TV interface (HTML/CSS/JS) — spatial D-pad navigation with clear focus states, layout that scales from 960×540 to 4K, film stills as live backdrops, and household memory stored on the device.
  • AI engine (Vercel serverless + Gemini) — Gemini classifies intent and extracts structured fields; film selection is deterministic against the catalog; Gemini then writes reasons using only catalog fields. It is instructed never to invent runtimes, awards or facts. The API key never leaves the server.
  • Resilience — if the AI is busy or unreachable, the engine retries, falls back to backup models, and finally to a keyword engine, so discovery always returns films.
  • GitHub Actions builds the APK on every push and publishes it to a release.

One engine serves both the web (engine.ver-de.com) and the TV.

Challenges

  • No device, no budget. I don't own a Fire TV or a smart TV, and buying one wasn't possible. I built and recorded on the Android TV emulator in Android Studio — the APK is a standard Fire OS app that sideloads onto Fire TV.
  • Infrastructure availability. My first planned AI provider didn't support billing in the Philippines, so I rebuilt the engine around Gemini.
  • A real PC with real limits. A full C: drive, virtualization disabled in the BIOS, and a model returning "high demand" errors the night before recording — each one solved with a fallback rather than a workaround that hides the problem.
  • Designing for a remote. Conversation usually means typing. On a TV it had to become a few honest choices that still feel personal.

What I learned

On a TV, intelligence should feel like fewer choices, not more features. Three films with a reason beat a wall of posters. And memory — knowing who is watching together — matters as much as what they asked for.

What's next

Install on a physical Fire TV device, add more VER-DÉ films with full runtimes, open the catalog to other independent filmmakers, and explore Alexa voice requests as a second way to ask.

A VER-DÉ Film · Music by Green Ralph · Nabua, Camarines Sur

Built With

Share this project:

Updates

Submission history