30-Second Read

Key Decisions

Reframed the product as 'help me decide what to watch,' not another catalog — and made a 4-action feedback model (Pass / Interested / Save / Watched) the single signal driving exclusion, preference weight, and queue refresh.

AI · Data Logic

Built a transparent rule-based ranking engine first — 7 signals + 3 diversity guardrails over a 6-table schema — deliberately deferring ML until real feedback exists, with the schema designed from day one to train an embedding / collaborative-filtering layer later.

Result

Built a 23-screen interactive prototype covering the full onboarding → recommendation → feedback → saved-library loop, making the core decision flow testable end-to-end.

What I Learned

Designed the recommendation evaluation around acceptance, swipe-completion, and save/watched ratios — and learned to treat the feedback schema as the real product surface, since both today's rules and tomorrow's model live or die on signal quality.

Coming SoonMobile app · Recommendation system · Media discovery · Personalized decisioning

VibeWatch

A recommendation-first media discovery app that helps you find your next watch in a few swipes.

Role

AI Product Manager · Product Owner

Timeline

Mar 2026 – Present

Core Problem

As streaming catalogues grow, the problem is no longer a lack of content but the difficulty of deciding what to watch now. Traditional flows rely on long lists, generic rows, ratings, and manual comparison, creating decision fatigue that rarely matches the viewer's current mood. VibeWatch turns watch selection into a faster, lighter, more personalized flow: pick a content type and vibe, then swipe while the system learns.

Target Users

Viewers tired of endless scrolling who want fast, personalized picks — especially people watching across multiple platforms who struggle to decide between movies, TV, variety, anime, or documentaries on a given night.

Key Results

—Defined four core feedback actions (Pass / Interested / Save / Watched) that turn explicit user actions into preference, exclusion, and library state.
—Built a 23-screen React Native prototype covering the full onboarding → recommendation → feedback-capture → saved-library loop, making the core 'help me decide what to watch' flow testable end-to-end (90 files, 12 test files).
—Planned 6 core Supabase/PostgreSQL tables across users, preferences, content metadata, external IDs, behavior logs, and recommendation sessions.
—Closed the product loop from PRD, design system, IA, and navigation to a mobile prototype — framing VibeWatch as a 'help me decide what to watch' tool, not a streaming player.

⌥GitHub↗