A television tracker for exactly two users: my wife and me.
A TV tracker built for two — one shared catalog, free-text notes the AI logs correctly, and weekly picks for both of us.
What should we watch next? Did we ever finish that season? Where are we watching that — Netflix, Hulu, Prime? Ugggghhhhhh. Those are just examples — the kinds of questions every couple ends up asking the TV. The tracking tools that exist are built around one person's taste — none of them model the thing that actually matters on a Tuesday night: what do we both want?
BG Watch is built for two from the schema up. One shared catalog; two independent watch histories and sets of preferences; and mutual picks — the shows we'd both enjoy — surfaced with a badge. Mark an episode watched while we're watching together, and it's recorded for both of us.
The AI does the tedious parts. Type — or just speak — a free-text note like "watched the first two episodes of Slow Horses last night" and the right episodes get logged: an auto-tagger parses the note, matches it to the show and episodes, and records it with a confidence score attached, so a bad guess never silently corrupts the history. Every Sunday, a recommendation engine reads both taste profiles and scores candidates into a short weekly what-to-watch-next list.
And it puts everything in one place. Ratings arrive automatically; every show links out to its trailer; and the drill-down no remote provides is built in — from a show to its per-season cast, to any actor, to "who have we seen them in before — and is any of it already in our catalog?" Add the weekly suggestions, and it's one place instead of five apps.
It's a small app, but it was fun to build — and it uses the same rigor as the big ones: the same test-driven agent workflow, the same automated browser QA, seventy test files for a two-person tool. Unlike the enterprise work, it can be shown end to end.
Shared catalog, two histories, mutual picks badged.
Free text in; show, season, and episode matched automatically.
Weekly AI picks scored against both taste profiles.
Show → cast → "what else are they in that we already have?"
All screenshots captured from a demo copy seeded with sample data — no real viewing history.