0-to-1 Product · Consumer Mobile App

SwipeFlix

Reimagining movie discovery through swipe-based personalization

A full-stack mobile product built, owned, and shipped by Vibin Das — combining behavioral AI, real-time personalization, and habit-forming UX to solve one of streaming's oldest problems: finding something to watch.

Live Prototype iOS & Android React Native · Supabase 0-to-1 Product AI Recommendation Engine

Explore the case → PM skills demonstrated

The problem

Streaming is broken for discovery

Despite thousands of titles across dozens of platforms, users spend more time browsing than watching. SwipeFlix was born to fix this.

😵

Decision Fatigue

Infinite scroll UIs create overwhelming choice paralysis. Users spend 20+ minutes deciding — and often give up entirely. More options shown means harder decisions made.

🎯

Weak Personalization

Existing algorithms optimize for platform engagement, not user satisfaction. They surface what's trending, not what you'll personally love — leading to repeated, irrelevant suggestions.

🧩

Fragmented Ecosystem

Your watchlist is scattered across Netflix, Prime, Hotstar, Apple TV, and more. No unified discovery layer exists to help you find and track content across all platforms in one place.

"Finding what to watch should be as fun as watching it." — SwipeFlix Product Vision

The product manager

Built, owned, and shipped by Vibin

SwipeFlix demonstrates full product ownership — from blank canvas to live mobile app. Every decision reflects deliberate PM thinking.

Vibin Das

Sr. PM · Amazon

Full-stack product ownership — concept to App Store

Vibin conceptualised the product, defined the problem space, designed the recommendation logic, shaped the UX, collaborated on the technical architecture, and drove it to a live mobile product. Every decision reflects deliberate PM thinking: what to build, what to cut, and why.

Product Strategy

Defined MVP scope, identified user pain points, designed engagement loops, and prioritised the roadmap with deliberate ruthlessness.

User Experience

Structured onboarding flows, defined swipe interaction logic, reduced friction to near-zero, and built retention-first experiences.

Data & Personalization

Designed the preference scoring system, defined the recommendation weighting model, and introduced exploration logic to prevent stagnation.

Technical Collaboration

Worked across frontend, backend, and analytics. Personally debugged critical auth and schema issues — shipping fixes, not filing tickets.

Core features

Every feature solves a real user problem

Feature decisions driven by user behavior, engagement loop design, and validated hypotheses — not feature requests.

👆

Swipe Discovery Engine

Right = Watchlist. Left = Skip. Up = Watched. Three gestures replace thousands of micro-decisions — reducing time-to-decision dramatically and creating a habit loop through kinetic satisfaction. Transforms content paralysis into instant commitment.

UX SimplificationHabit LoopEngagement

🧠

Real-Time Preference Engine

Every swipe updates the user's preference profile across genres, languages, eras, and content types in real time. The system doesn't just collect data — it acts on it immediately, making each session smarter than the last. This closed feedback loop is the core of the product's stickiness.

PersonalizationBehavioral DataRetention

🌱

Cold Start Problem — Solved

New users get poor recommendations everywhere because there's no history. SwipeFlix uses lightweight onboarding questions to seed initial preferences, then rapidly calibrates after the first few swipes. Users feel understood from session one — not session ten.

Onboarding DesignML Cold StartActivation

🎲

Exploration Picks (Anti-Filter-Bubble)

Pure personalization traps users in echo chambers. SwipeFlix deliberately injects an "Exploration Pick" every 5 swipes — surfacing content outside the user's usual preferences. This strategic serendipity preserves novelty, prevents stagnation, and drives long-term retention.

Filter Bubble PreventionLong-term Retention

📚

Unified Library & Watchlist

A single home for everything: Watchlist, Watched, and Skipped. Filter by status to turn scattered decisions into a coherent personal catalog — reducing the anxiety of "what do I have saved?" and enabling intentional return visits and re-engagement.

Retention FeatureCatalog Management

📊

Admin Dashboard

Real-time analytics on user growth, swipe activity, platform popularity, and content ingestion. Role-based access controls. Automated TMDB sync with zero manual content work. Vibin designed internal tooling with the same rigor as consumer features — because great PMs know internal tools directly enable scale.

Internal ToolingAnalyticsScalability

AI & intelligence

AI isn't a feature — it's the product

Vibin designed SwipeFlix with AI embedded at the product layer. Every interaction is a data point; every data point makes the product smarter.

SwipeFlix's recommendation intelligence is a custom-built weighted scoring AI system that learns from behavioral signals in real time. Vibin designed the entire model architecture as a PM — defining the signals, weights, feedback loops, and exploration strategy.

🎯

Behavioral Signal Collection

Every swipe (right, left, up) is a labeled training example. The system tracks implicit signals — swipe direction, recency, and genre clustering — to build a rich user preference vector updated in real time via Supabase.

⚖️

Weighted Scoring Model

Vibin personally designed the ranking formula: Preference Match (60%) + Popularity (20%) + Freshness (10%) + Serendipity (10%). A multi-objective optimization that balances personalization against discovery — a classic PM tradeoff.

🌱

Cold Start Intelligence

Most AI systems fail new users. SwipeFlix uses onboarding-seeded preference initialization to give the model a starting point, then rapidly calibrates within the first session — without requiring a long questionnaire.

🛡️

Self-Healing Architecture

Auto-detects and cleans "metadata bloat" in auth sessions (JWT limit awareness). Parallel data fetching with Promise.all keeps recommendations fresh with sub-700ms load times — AI that doesn't sacrifice performance.

Phase 2 AI roadmap — designed by Vibin

AI "What should I watch tonight?" LLM assistant

Mood-based discovery with NLP intent parsing

Social graph recommendations from friends' tastes

Smart notifications for new-release preference matching

Streaming subscription sync & cross-platform availability

Collaborative filtering across anonymized user cohorts

Recommendation logic

The algorithm Vibin designed

A custom weighted ranking system designed by a PM — demonstrating deep technical fluency alongside product thinking.

Recommendation Scoring Weights

Preference Match60%

Popularity Signal20%

Freshness / Recency10%

Serendipity / Exploration10%

Exploration Pick Injection

Every 5th swipe is deliberately replaced with an out-of-preference "Exploration Pick" — a PM-designed nudge to prevent filter bubbles and maintain discovery novelty.

📥 Real-Time Preference Updates

User preference vectors update synchronously after every swipe via Supabase Realtime. No batch processing — the model reflects the most recent interaction within milliseconds.

🔑 Multi-Dimensional Profiling

Preferences tracked across: Genre · Language · Content Type · Release Era · Popularity Tolerance. This 5-dimensional vector makes recommendations highly specific — not just "you like thrillers," but "you like 90s Korean thrillers."

⚡ Performance Architecture

Parallel data fetching using Promise.all loads user profiles and swipe actions concurrently — reducing startup latency by ~50% (500–700ms total). Safety timeouts prevent hangs on slow networks.

🛡️ Anti-Bloat Metadata Design

JWT tokens have a 100KB limit. Vibin architected the fix: moving user action data into normalized Supabase tables with real-time sync — solving a production-critical bottleneck most PMs never know exists.

Product decisions & tradeoffs

Every "no" is as important as every "yes"

The mark of a great PM is clarity on tradeoffs. These decisions show how Vibin thinks when competing priorities collide.

Why swipe over lists?

Traditional list UIs create cognitive overload. Swipe interfaces force a binary decision — reducing cognitive load to near-zero, creating momentum, and generating faster feedback loops. The tradeoff: less browsing context per title, offset by higher decision velocity.

Why mix exploration with personalization?

Pure personalization is a long-term retention risk. It creates filter bubbles and eventually bores users. Deliberate serendipity (10% weight + every-5th-swipe injection) preserves novelty. Short-term relevance sacrificed for long-term engagement.

Why mobile-first?

Discovery behavior is mobile-native: commutes, breaks, pre-sleep browsing. Desktop would serve a different (smaller) use case. MVP focus on mobile maximized core-user alignment. Web excluded from v1 intentionally to avoid scope creep.

Why lightweight onboarding?

Long questionnaires kill activation. Vibin chose a 3-question onboarding that seeds preferences "good enough" — then lets the swipe algorithm calibrate rapidly. Activation rate prioritized over initial recommendation quality.

Why build an Admin Dashboard?

Internal tooling is PM infrastructure. Without it, scaling requires expensive human oversight. Building real-time analytics, content ingestion monitoring, and RBAC in v1 creates a scalable operational backbone — not a nice-to-have.

Why TMDB over a proprietary catalog?

Building a proprietary catalog would take months and millions. TMDB provides a battle-tested global API with 1M+ titles. Tradeoff: third-party dependency. Offset by automated sync (zero manual overhead) and massive time-to-market acceleration.

Challenges solved

Hard problems, sharp solutions

The challenges Vibin identified and solved in SwipeFlix mirror the complexity of enterprise-scale product problems — in a consumer context.

Recommendation Cold Start

Without behavioral history, new users receive generic, irrelevant recommendations — destroying first-session satisfaction and activation rates.

✓ Solved via onboarding preference seeding + rapid real-time calibration within first 10 swipes

JWT Metadata Bloat (Session Limits)

Storing user swipe actions in JWT auth tokens approached the 100KB limit — causing authentication failures in production for power users.

✓ Migrated user actions into normalized Supabase tables with real-time sync — eliminating the token size constraint entirely

Slow Startup — Sequential Data Loading

Initial auth flow loaded user profile and action history sequentially — creating 1200ms+ startup delays that undermine the "snappy" product feel.

✓ Refactored to concurrent loading with Promise.all — reducing startup to 500–700ms, a ~50% performance improvement

Database Schema Mismatch (Login Bug)

The app queried for a "role" column that didn't exist in the profiles table — causing silent authentication failures with no visible error. Users simply couldn't log in.

✓ Debugged via diagnostic script, updated query to only fetch existing columns, implemented email-based role overrides — full resolution verified

Discovery Fatigue & Filter Bubbles

Strong personalization without intervention creates echo chambers — novelty drops, sessions shorten, and eventually churn accelerates. The paradox of over-personalization.

✓ Algorithmic Exploration Picks (every 5th swipe) inject deliberate serendipity to preserve novelty and long-term retention

Metrics framework

How success is measured

Vibin defined a full measurement framework before building. Every feature is tied to a metric; every metric rolls up to the North Star.

North Star Metric

Successful Watches Started from Recommendations

Not "swipes per session." Not "DAU." The metric that proves the product actually works.

Acquisition

Signup conversion rate

Onboarding completion rate

Time-to-first-swipe

Engagement

Swipes per session

Sessions per week

Watchlist adds

Retention

D7 retention rate

D30 retention rate

Repeat discovery sessions

Recommendation Quality

Recommendation acceptance rate

Watchlist → Watch conversion

Time-to-first-watch

Product Health

Startup latency (target <700ms)

Auth success rate

Metadata bloat events

Admin & Ops

TMDB sync latency

Content catalog freshness

Content ingestion uptime

Technical execution

Built with production-grade architecture

Technical awareness is a PM superpower. Vibin collaborated directly on architecture decisions — not just requirements docs.

Frontend

React Native + Expo

Routing

Expo Router

State Mgmt

React Context API

Animations

Reanimated + Haptics

Backend / DB

Supabase (PostgreSQL)

Auth

Supabase Auth

Infrastructure

Vercel Edge Functions

Content Data

TMDB API

⚡ 50% Faster Startup

Parallel data fetching reduced initial load from 1200ms+ to 500–700ms

🛡️ Self-Healing Auth

Auto-detection and purging of JWT metadata bloat prevents silent auth failures

🔄 Zero Manual Content

Automated TMDB sync keeps the catalog fresh with no human intervention needed

PM skills demonstrated

What SwipeFlix proves about Vibin

This isn't just a side project — it's a demonstration of every skill hiring managers look for in a senior product leader.

Vibin built SwipeFlix the same way he builds Amazon's supply chain products — with rigorous problem framing, deliberate tradeoff decisions, metrics-first thinking, and cross-functional execution discipline. The stack is different; the PM muscle is identical.

0-to-1 Product Development

Took SwipeFlix from blank canvas to live App — defining the problem, scoping the MVP, and driving every decision to launch.

AI / ML Product Thinking

Designed the recommendation algorithm, weighted scoring model, cold start solution, and exploration logic — without a data science team.

Personalization Systems

Built a real-time behavioral preference engine tracking 5 dimensions of user taste — updated synchronously on every interaction.

Engagement Loop Design

Architected swipe mechanics, exploration injection, and the watchlist as interconnected retention and re-engagement loops.

Technical PM Depth

Personally debugged schema mismatches, JWT limits, and sequential loading bugs — shipping fixes, not filing tickets.

Metrics-First Thinking

Defined the North Star metric and full measurement pyramid before writing a single line of product code.

UX Simplification

Reduced an overwhelming multi-platform browsing problem to three swipe gestures — eliminating decision fatigue by design.

Internal Tooling Strategy

Built the Admin Dashboard as a v1 priority — treating operational infrastructure as a product, not an afterthought.

Product Tradeoff Clarity

Six deliberate tradeoff decisions documented — showing why every "no" was as deliberate as every "yes."

Phase 2 roadmap

What Vibin would build next

A PM who can articulate the next three versions of a product understands where the real value lies.

AI "What should I watch tonight?" assistant — Conversational discovery powered by LLM with context from swipe history, mood, and available time. The natural evolution of the recommendation engine.

AI/LLM

Social recommendations & shared watchlists — See what friends are watching, build shared lists for movie nights. Turns solo discovery into a social product — unlocking viral growth loops.

Social

Streaming subscription sync — Connect Netflix, Prime, Hotstar, and others to surface only titles available on your platforms. Solving the ecosystem fragmentation that motivated SwipeFlix.

Integration

Mood-based discovery with NLP — "I'm in the mood for something funny and short" parsed into recommendation filters. Makes intent-based discovery as easy as talking to a friend.

NLP

Smart notifications for new-release matching — Alert users when a title matching their taste profile drops on a platform they own. Turns the app into a proactive discovery companion.