Concept · Interactive Prototype

Netflix Personalization, rethought around your day

A speculative product mockup: what recommendations could feel like if your streaming service understood your mood, schedule, and energy — not just your watch history.

RoleDesign & Prototype · 2024

LifeContext

What if Netflix knew how your day went?

A personalization concept that fuses lifestyle signals — activity, mood, energy, time — into the recommendation pipeline. Watch the homepage transform as life context changes.

Choose a Persona

🏃
Marathon Runner
Training for a marathon, high activity, structured schedule
👶
New Parent
Sleep-deprived, limited time, needs comfort
💻
Remote Worker
WFH, blurred work/life, varied energy throughout day
🌍
Weekend Explorer
Active social life, loves trying new things, mood-driven

Time of Day

Current Mood

Because You're Feeling Pumped

Drive to Survive
99% match
Drive to Survive
Sports Doc · 2019
Cobra Kai
99% match
Cobra Kai
Action Comedy · 2018
The Last Dance
99% match
The Last Dance
Sports Doc · 2020
Tour de France: Unchained
99% match
Tour de France: Unchained
Sports Doc · 2023
Untold
99% match
Untold
Sports Doc · 2021
Narcos
94% match
Narcos
Crime Drama · 2015
Peaky Blinders
94% match
Peaky Blinders
Crime Drama · 2013
Ozark
93% match
Ozark
Crime Drama · 2017

For Your Morning

The Witcher
92% match
The Witcher
Fantasy · 2019
Money Heist
88% match
Money Heist
Thriller · 2017
Breaking Bad
87% match
Breaking Bad
Drama · 2008
Stranger Things
87% match
Stranger Things
Sci-Fi · 2016
Squid Game
87% match
Squid Game
Thriller · 2021
You
87% match
You
Thriller · 2018
Lupin
82% match
Lupin
Thriller · 2021
Our Planet
78% match
Our Planet
Documentary · 2019

Your Lifestyle Picks

Chef's Table
77% match
Chef's Table
Documentary · 2015
Somebody Feed Phil
77% match
Somebody Feed Phil
Food Travel · 2018
Ted Lasso
76% match
Ted Lasso
Comedy · 2020
The Crown
74% match
The Crown
Drama · 2016
Wednesday
74% match
Wednesday
Comedy Horror · 2022
Mindhunter
74% match
Mindhunter
Crime Drama · 2017
The Social Dilemma
74% match
The Social Dilemma
Documentary · 2020
All Quiet on the Western Front
74% match
All Quiet on the Western Front
War Drama · 2022

Something Different

The Great British Baking Show
70% match
The Great British Baking Show
Reality · 2010
Bridgerton
70% match
Bridgerton
Romance · 2020
Never Have I Ever
70% match
Never Have I Ever
Comedy · 2020
Heartstopper
71% match
Heartstopper
Romance · 2022
Love Is Blind
71% match
Love Is Blind
Reality · 2020
Dark
71% match
Dark
Sci-Fi Thriller · 2017
Bojack Horseman
71% match
Bojack Horseman
Animated Comedy · 2014
The Office
72% match
The Office
Comedy · 2005

Why Lifestyle Context Matters

Netflix knows what you watch, but not who you are when you press play. The same person who craves adrenaline after a morning run wants pure comfort at 10pm. Current recommendation systems optimize for preference — collaborative filtering asks "what do people like you watch?" — but they miss the when and how of consumption.

The hypothesis is simple: recommendations that align with your physiological and emotional state will increase session satisfaction and reduce browse-to-play time. Instead of 20 minutes of scrolling, you find something that fits in under 60 seconds.

The key insight: mood is not preference. Preference is stable ("I like thrillers"). Mood is transient ("right now I need comfort"). The best systems should model both.

Architecture

External SignalsApple HealthStravaCalendarManual MoodTime of DayLocationContext EngineSignal processingState inferenceSignal FusionContext vectorsPreference vectorsWatch historyCollaborative filterEmbeddingsRe-RankingScore fusionRow assemblyHomePageNetflix FoundationModel + HistoryLifeContext Architecture PipelineLifestyle signals fused with content embeddings for context-aware re-ranking

How It Ships

Phase 1
Manual Mood Input
Low-friction "How are you feeling?" prompt at session start. Minimal UX change, maximum signal gain. Users opt-in with a single tap.
Phase 2
Passive Inference
Time-of-day and watch pattern analysis. No new data needed — just smarter use of existing signals. The system learns your circadian rhythm of content.
Phase 3
Device Integrations
Opt-in Apple Health, Strava, and Google Fit connections. On-device inference, differential privacy, and full user controls. Privacy by design from day one.
Success Metrics
↓ 30%
Browse-to-play time
↑ 15%
Session satisfaction
↑ 20%
Content diversity
↑ 5%
30-day retention

Feature Engineering

Raw signals like steps, heart rate, and sleep hours are transformed into a normalized context vector. Step count maps to an activity score (0–1), resting heart rate indicates recovery state, and sleep debt compounds over a 3-day window. Time of day is encoded as a cyclical feature using sine/cosine to capture the circular nature of the 24-hour clock.

context_vector = [activity_score, recovery_state, sleep_debt, time_sin, time_cos, mood_valence, mood_arousal]

Embedding Space

The user's state is represented as two vectors: a stable preference vector (learned from watch history via collaborative filtering) and a transient context vector (computed in real-time from lifestyle signals). These are combined via a learned attention mechanism that dynamically weights how much context should influence ranking. In high-confidence contexts (e.g., post-workout + high energy), context weight increases. In ambiguous states, the system falls back to preference-only ranking.

Multi-Armed Bandit

New context-content pairings need exploration. A Thompson Sampling approach balances exploitation (show high-confidence matches) with exploration (test uncertain pairings). The "Something Different" row explicitly serves this: it surfaces low-match titles where the system is uncertain, gathering signal on whether context-defying choices lead to engagement. This prevents the filter bubble problem that plagues pure optimization.

A/B Test Design

Control group receives standard collaborative filtering recommendations. Treatment group gets lifestyle-aware re-ranking with context signals.

Primary Metric
Qualified play rate — percentage of sessions where the user watches more than 5 minutes of the first selected title.
Guardrail Metric
Content diversity — ensuring the algorithm doesn't collapse into a narrow set of recommendations. Measured by distinct titles shown per session.

Key Insights

🎯
Mood ≠ Preference
Your favorite genre doesn't change. Your tolerance for it does. A thriller fan after a stressful day might need comedy. Preference is who you are — mood is how you are right now.
🕐
Context is Temporal
The same user is a different person at 7am vs 11pm. Morning-you craves energy. Evening-you craves comfort. A truly personal system must model this within-day variation.
🔮
Passive > Active
The best personalization doesn't ask. It infers. Requiring explicit input creates friction and drops off fast. The richest signals come from behavior, not questionnaires.
🎲
Serendipity Matters
Over-optimization kills discovery. Sometimes the best recommendation doesn't match your current state — it introduces you to something you didn't know you needed.

LifeContext is a concept exploration — not affiliated with Netflix. Poster images via TMDB.