Update index.html

index.html

@@ -418,25 +418,25 @@
   <div class="content" style="width:100%;max-width:1200px;margin:0 auto;">
     <div class="hero-grid">
       <div>
-        <div class="slide-tag">Decode the Tech · Season 2026</div>
+        <div class="slide-tag">Decode the Tech · Episode 4</div>
         <h1 class="hero">SIMPLE<br><span>INTERFACE.</span><br>HIDDEN SYSTEM.</h1>
         <div class="divider"></div>
         <p class="subtitle">Your Netflix homepage feels effortless. That feeling is the product. Behind it sits a layered machine learning system making hundreds of decisions before a single thumbnail loads.</p>
         <div class="hero-badges" style="margin-top:1.6rem;">
           <span class="badge badge-red">Machine Learning</span>
-          <span class="badge badge-white">Behavioral Signals</span>
-          <span class="badge badge-white">Multi-Stage Ranking</span>
+          <span class="badge badge-red">Behavioral Signals</span>
+          <span class="badge badge-red">Multi-Stage Ranking</span>
           <span class="badge badge-red">Personalization at Scale</span>
-          <span class="badge badge-white">Representation Systems</span>
+          <span class="badge badge-red">Representation Learning</span>
         </div>
         <div class="intro-reveal" style="margin-top:1.5rem;">
-          <p>Netflix <strong>personalizes</strong> more than which titles you see — it personalizes which rows appear, how titles are ordered within those rows, and even which thumbnail represents the same show to different viewers. <strong>This talk is not about movies. It is about how a familiar interface hides a layered machine learning system.</strong></p>
+          <p>Netflix <strong>personalizes</strong> more than which titles you see — it personalizes which rows appear, how titles are ordered within those rows, and even which thumbnail represents the same show to different viewers. <strong>A deep dive into the layered machine learning system behind a familiar interface.</strong></p>
         </div>
       </div>
       <div class="hero-stats">
-        <div class="hero-stat"><div class="num">250M+</div><div class="lbl">Global Subscribers</div></div>
-        <div class="hero-stat"><div class="num">~80%</div><div class="lbl">Content Discovered via Recommendations</div></div>
-        <div class="hero-stat"><div class="num">$30B+</div><div class="lbl">Annual Revenue (2024)</div></div>
+        <div class="hero-stat"><div class="num">325M+</div><div class="lbl">Paid Memberships Worldwide</div></div>
+        <div class="hero-stat"><div class="num">80%+</div><div class="lbl">Viewing Discovered via Recommendations</div></div>
+        <div class="hero-stat"><div class="num">$45.2B</div><div class="lbl">Annual Revenue (FY2025)</div></div>
         <div class="hero-stat"><div class="num">190+</div><div class="lbl">Countries · 60+ Languages</div></div>
       </div>
     </div>
@@ -510,16 +510,18 @@
 <!-- ══════════════ SLIDE 3: WHY NETFLIX ══════════════ -->
 <section class="slide" id="s3">
   <div class="content" style="width:100%;max-width:1200px;margin:0 auto;">
-    <div class="slide-tag">Justifying the Case Study</div>
+    <div class="slide-tag">Why This System Matters</div>
     <h2 class="sec">WHY NETFLIX<br><span>IS THE RIGHT EXAMPLE</span></h2>
-    <p class="subtitle">Not just because it's popular — because recommendation is structurally central to how the product works, and Netflix has published unusually useful technical material to study.</p>
+    <p class="subtitle">Recommendation is not a side feature here — it shapes discovery, presentation, ranking, and long-term retention across the entire product.</p>
     <div class="why-grid">
-      <div class="why-card"><div class="why-icon">🔭</div><div class="why-title">Recommendation Drives Discovery</div><div class="why-desc">With a catalog of hundreds of thousands of titles, no user can browse manually. The system is not a feature on top of the product — it IS how users navigate the catalog. Without it, most content is invisible.</div><span class="why-metric">Discovery problem at scale</span></div>
-      <div class="why-card"><div class="why-icon">🧩</div><div class="why-title">Multi-Level Personalization</div><div class="why-desc">Netflix personalizes beyond item ranking. It personalizes which rows appear, how titles are ordered within those rows, and how each title is visually represented. That makes it a richer system to study than a simple ranked list.</div><span class="why-metric">Row · Title · Artwork layers</span></div>
-      <div class="why-card"><div class="why-icon">📚</div><div class="why-title">Unusually Transparent Documentation</div><div class="why-desc">Through its Help Center, Tech Blog, and Research pages, Netflix has published detailed explanations of its signals, ranking logic, and what the system explicitly does not use — enabling a more precise study than most commercial systems allow.</div><span class="why-metric">Publicly documented system</span></div>
-      <div class="why-card"><div class="why-icon">📊</div><div class="why-title">The Data Flywheel</div><div class="why-desc">More users → more behavioral signals → better models → better recommendations → more retained users. This loop has been compounding for 20+ years, creating a structural moat that is difficult to replicate quickly.</div><span class="why-metric">Self-reinforcing advantage</span></div>
-      <div class="why-card"><div class="why-icon">🔬</div><div class="why-title">Scientific A/B Testing Culture</div><div class="why-desc">Netflix runs thousands of A/B tests per year. Every feature — row order, thumbnail variant, algorithm configuration — is tested before shipping. This institutional rigor is genuinely rare in entertainment companies.</div><span class="why-metric">Data-driven at every layer</span></div>
-      <div class="why-card"><div class="why-icon">🤖</div><div class="why-title">System Evolution is Traceable</div><div class="why-desc">From the 2009 Netflix Prize to today's foundation models, the public record of Netflix's system evolution spans nearly two decades — making it an unusually instructive case study in how industrial ML systems mature.</div><span class="why-metric">From Prize era to FM-Intent</span></div>
+      <div class="why-card"><div class="why-icon">🔭</div><div class="why-title">Discovery at Catalog Scale</div><div class="why-desc">With an enormous catalog, browsing alone does not work. Recommendation is the product layer that makes the catalog navigable and watchable at all.</div><span class="why-metric">Discovery problem at scale</span></div>
+      <div class="why-card"><div class="why-icon">🧩</div><div class="why-title">Layered Personalization</div><div class="why-desc">Netflix personalizes rows, the titles inside those rows, their order, and even the visual representation of the same title for different members.</div><span class="why-metric">Rows · Titles · Artwork</span></div>
+      <div class="why-card"><div class="why-icon">📚</div><div class="why-title">Public Technical Material</div><div class="why-desc">Netflix has published unusually useful explanations across its Help Center, Tech Blog, and Research site, making the system easier to decode than most commercial recommenders.</div><span class="why-metric">Rare public visibility</span></div>
+      <div class="why-card"><div class="why-icon">⚙️</div><div class="why-title">Multiple Models, Not One</div><div class="why-desc">Netflix describes specialized recommendation models for different homepage surfaces and use cases, which makes it a better real-world example than a single ranked-list demo.</div><span class="why-metric">Production system complexity</span></div>
+      <div class="why-card"><div class="why-icon">🖼️</div><div class="why-title">Presentation Is Personalized Too</div><div class="why-desc">The system does not stop after choosing a title. It also chooses how that title is shown, including which artwork is most likely to attract the right viewer.</div><span class="why-metric">Recommendation beyond ranking</span></div>
+      <div class="why-card"><div class="why-icon">⏱️</div><div class="why-title">Latency Meets Personalization</div><div class="why-desc">Netflix must make all of these decisions fast enough for a homepage to feel instant, which turns recommendation into both an ML problem and a systems problem.</div><span class="why-metric">Millisecond serving constraints</span></div>
+      <div class="why-card"><div class="why-icon">🧪</div><div class="why-title">Experimentation Culture</div><div class="why-desc">Rows, ranking logic, and visual treatments are continuously evaluated, which makes the system a strong example of ML tied directly to product experimentation.</div><span class="why-metric">Measured product iteration</span></div>
+      <div class="why-card"><div class="why-icon">🚀</div><div class="why-title">Clear Evolution Path</div><div class="why-desc">Netflix gives a rare view of how recommenders evolve — from collaborative filtering era ideas to deep learning, contextual personalization, and foundation-model direction.</div><span class="why-metric">From Prize era to foundation models</span></div>
     </div>
   </div>
 </section>
@@ -530,56 +532,56 @@
   <div class="content" style="width:100%;max-width:1200px;margin:0 auto;">
     <div class="slide-tag">Decode the Tech · Inputs to the System</div>
     <h2 class="sec">WHAT GOES<br><span>INTO THE SYSTEM</span></h2>
-    <p class="subtitle">Netflix's recommendations are shaped by four categories of signals — behavioral, collaborative, content, and contextual. Notably, demographic data like age and gender are not part of the decision process.</p>
+    <p class="subtitle">Netflix's recommendations are shaped by four categories of signals — interaction, collaborative, content, and context. Its public explanation also states that demographic attributes like age and gender are not part of the recommendation decision.</p>
     <div class="signals-layout">
       <div>
         <div class="signal-categories">
           <div class="sig-cat">
-            <div class="sig-cat-label">Behavioral</div>
-            <div class="sig-cat-title">Your Interaction History</div>
-            <div class="sig-cat-items"><span>Watch history</span>, <span>completion rate</span>, <span>thumbs ratings</span>, <span>pause behavior</span>, <span>rewind patterns</span>, <span>hover time on thumbnails</span>, <span>binge speed</span>, <span>scroll depth without clicking</span></div>
+            <div class="sig-cat-label">Interaction</div>
+            <div class="sig-cat-title">What You Actually Do</div>
+            <div class="sig-cat-items"><span>Watch history</span>, <span>completion rate</span>, <span>rewatches</span>, <span>thumbs feedback</span>, <span>searches</span>, <span>browse clicks</span>, <span>watch duration</span>, <span>session-to-session recency</span></div>
           </div>
           <div class="sig-cat">
             <div class="sig-cat-label">Collaborative</div>
             <div class="sig-cat-title">Patterns from Similar Members</div>
-            <div class="sig-cat-items">What users with <span>similar taste profiles</span> watched, completed, and rated highly — used to recommend content you haven't seen but users like you have enjoyed</div>
+            <div class="sig-cat-items">What members with <span>similar taste patterns</span> watched, finished, replayed, or ignored — useful for recommending titles you have never seen before</div>
           </div>
           <div class="sig-cat">
             <div class="sig-cat-label">Content</div>
-            <div class="sig-cat-title">Title Metadata</div>
-            <div class="sig-cat-items"><span>Genre</span>, <span>actors and directors</span>, <span>categories and tags</span>, <span>release year</span>, <span>language</span>, <span>pacing and format</span></div>
+            <div class="sig-cat-title">What the Title Is</div>
+            <div class="sig-cat-items"><span>Genre</span>, <span>cast and creators</span>, <span>release year</span>, <span>language</span>, <span>maturity profile</span>, <span>format</span>, <span>metadata tags</span>, <span>learned title representations</span></div>
           </div>
           <div class="sig-cat">
-            <div class="sig-cat-label">Contextual</div>
-            <div class="sig-cat-title">When, Where, and How You Watch</div>
-            <div class="sig-cat-items"><span>Device type</span> (TV vs. phone vs. laptop), <span>time of day</span>, <span>preferred language</span>, <span>recent session behavior</span></div>
+            <div class="sig-cat-label">Context</div>
+            <div class="sig-cat-title">When the Request Happens</div>
+            <div class="sig-cat-items"><span>Device type</span>, <span>time of day</span>, <span>member language</span>, <span>current session state</span>, <span>recent actions</span>, <span>request-time context</span></div>
           </div>
         </div>
         <div class="not-used">
           <div class="not-used-label">⚠ Not used in recommendations</div>
-          <div class="not-used-desc">According to Netflix's own explanation of its system, <strong style="color:var(--text);">age and gender are not part of the decision-making process.</strong> The system relies on behavioral, collaborative, content, and contextual signals — not demographic attributes.</div>
+          <div class="not-used-desc">According to Netflix's public explanation, <strong style="color:var(--text);">age and gender are not used as recommendation inputs.</strong> The emphasis is on behavior, title information, similar-member patterns, and viewing context.</div>
         </div>
       </div>
       <div class="signal-story">
-        <div class="ss-title">WHAT NETFLIX <span>INFERS</span><br>FROM YOUR BEHAVIOR</div>
+        <div class="ss-title">WHAT NETFLIX <span>INFERS</span><br>FROM YOUR SIGNALS</div>
         <div class="signal-scenario">
-          <div class="sc-action">You rewound that one scene three times</div>
-          <div class="sc-inference">Netflix infers: this type of scene — this actor, this tension style, this narrative beat — generates unusually high engagement for you. Similar content gets promoted in your next session.</div>
-          <div class="sc-tag">SIGNAL → INFERENCE → RANK ADJUSTMENT</div>
+          <div class="sc-action">You finish three tense crime episodes in one night</div>
+          <div class="sc-inference">The system infers short-term intent, high current appetite, and strong affinity for that viewing mode. Similar titles and rows gain weight in the next request.</div>
+          <div class="sc-tag">SESSION INTENT → RE-RANK</div>
         </div>
         <div class="signal-scenario">
-          <div class="sc-action">You abandoned a show after episode 2</div>
-          <div class="sc-inference">Netflix infers: slow-burn format may not suit your viewing style. Future ranking de-weights slow-paced shows in your recommendations — even ones you've never seen.</div>
-          <div class="sc-tag">IMPLICIT NEGATIVE SIGNAL → PROFILE UPDATE</div>
+          <div class="sc-action">You stop a slow series early but fully finish fast-paced thrillers</div>
+          <div class="sc-inference">The model updates your profile away from slow-burn pacing and toward tighter, faster narrative patterns — even across unseen titles.</div>
+          <div class="sc-tag">IMPLICIT NEGATIVE + POSITIVE SIGNALS</div>
         </div>
         <div class="signal-scenario">
-          <div class="sc-action">You scroll past 40 thumbnails without clicking</div>
-          <div class="sc-inference">Netflix infers: the current recommendation set missed the mark. The absence of a click is itself a signal — and can trigger a homepage refresh or diversity injection.</div>
-          <div class="sc-tag">INACTION IS ALSO A SIGNAL</div>
+          <div class="sc-action">You browse on phone at lunch and on TV at night</div>
+          <div class="sc-inference">Context changes the recommendation surface. The same member can receive different rows, orderings, and emphasis depending on device, time, and recent session behavior.</div>
+          <div class="sc-tag">REQUEST-TIME CONTEXT MATTERS</div>
         </div>
       </div>
       <div class="signals-footer">
-        <strong>The key insight:</strong> Netflix collects both the signals you consciously send (ratings, searches) and the signals you never thought to send — your hesitations, your timing, your device context. Newer interactions carry more weight than older ones; the system continuously re-weights your profile based on recency.
+        <strong>The key insight:</strong> Netflix learns from both explicit actions and implicit behavior. The system combines long-term taste, short-term session intent, title metadata, and request-time context to decide what to surface next.
       </div>
     </div>
   </div>
@@ -590,43 +592,43 @@
   <div class="content" style="width:100%;max-width:1200px;margin:0 auto;">
     <div class="slide-tag">System Design · What the System Optimizes For</div>
     <h2 class="sec">WHAT IS THE SYSTEM<br><span>ACTUALLY TRYING TO DO?</span></h2>
-    <p class="subtitle">Before understanding the architecture, you need to understand the objectives. Netflix's recommendation system is not simply predicting clicks — it is balancing multiple competing goals simultaneously.</p>
+    <p class="subtitle">Before architecture, understand the objective function. A modern recommender does not optimize for clicks alone — it balances satisfaction, speed, freshness, and long-term value under product constraints.</p>
     <div class="objectives-grid">
       <div class="obj-card featured">
         <div class="obj-icon">🎯</div>
         <div class="obj-title">Relevance</div>
-        <div class="obj-desc">Surface titles you are likely to genuinely enjoy — matching your established taste profile based on behavioral history, collaborative signals, and content metadata. The most fundamental goal.</div>
-        <div class="obj-note">PRIMARY · Match taste profile to catalog</div>
+        <div class="obj-desc">Match each member with titles they are genuinely likely to value, using interaction history, similar-member patterns, metadata, and learned representations instead of a single rule.</div>
+        <div class="obj-note">PRIMARY · Personalized utility estimation</div>
       </div>
       <div class="obj-card featured">
         <div class="obj-icon">⚡</div>
         <div class="obj-title">Discovery Speed</div>
-        <div class="obj-desc">Help you find something to watch quickly. Netflix's internal research indicates that users who cannot find something to watch within a short window are significantly more likely to leave the platform. The homepage layout and row ordering are optimized for fast discovery — not exhaustive browsing.</div>
-        <div class="obj-note">CRITICAL · Time-to-engagement matters</div>
+        <div class="obj-desc">Reduce time-to-first-play. A strong homepage gets members to a good decision quickly, so ranking, row ordering, and presentation all help compress search time.</div>
+        <div class="obj-note">CRITICAL · Lower decision friction</div>
       </div>
       <div class="obj-card">
         <div class="obj-icon">🔄</div>
-        <div class="obj-title">Freshness and Recency</div>
-        <div class="obj-desc">Newer behavioral signals carry more weight than older ones. A show you watched last week affects your recommendations more than something you watched six months ago. The system continuously re-weights your profile based on recency — preventing stale recommendations.</div>
-        <div class="obj-note">Recent interactions supersede old behavior</div>
+        <div class="obj-title">Freshness and Adaptation</div>
+        <div class="obj-desc">The model should react quickly when a member's taste shifts. Recent actions, session intent, and request-time context help prevent stale recommendations from dominating the page.</div>
+        <div class="obj-note">Fast profile updates from new behavior</div>
       </div>
       <div class="obj-card">
         <div class="obj-icon">🌱</div>
-        <div class="obj-title">Long-Term Retention</div>
-        <div class="obj-desc">A system that only optimizes for immediate clicks can narrow your perceived catalog over time. Netflix also optimizes for long-term engagement — which sometimes means showing you content slightly outside your comfort zone rather than repeating the same genre loop.</div>
-        <div class="obj-note">Beyond click prediction → sustained value</div>
+        <div class="obj-title">Long-Term Satisfaction</div>
+        <div class="obj-desc">The best system is not the one that only gets the next click. It should broaden the member's useful catalog over time and improve the chance they return tomorrow, next week, and next month.</div>
+        <div class="obj-note">Beyond immediate engagement</div>
       </div>
     </div>
     <div class="explore-exploit">
-      <div class="ee-label">The Core Tension · Exploration vs. Exploitation</div>
+      <div class="ee-label">Exploration vs. Exploitation</div>
       <div class="ee-cols">
         <div>
-          <div class="ee-col-title ee-exploit">⚡ Exploitation — Play it safe</div>
-          <div class="ee-desc">Show the user what the model is most confident they'll enjoy right now — proven genres, familiar formats, high-rated similar titles. Maximizes short-term engagement but risks creating a genre rut over time if left unchecked.</div>
+          <div class="ee-col-title ee-exploit">⚡ Exploitation — Maximize near-term confidence</div>
+          <div class="ee-desc">Rank what the system already believes is most likely to work: familiar genres, reliable franchises, strong collaborative matches, and high-confidence titles for the current session.</div>
         </div>
         <div>
-          <div class="ee-col-title ee-explore">🌱 Exploration — Take a small bet</div>
-          <div class="ee-desc">Intentionally include content slightly outside the established profile — a new genre, an unfamiliar format, an under-watched title. Most bets don't land. But the ones that do expand the taste model and increase long-term retention. This is sometimes discussed in the context of bandit-style approaches in personalization systems.</div>
+          <div class="ee-col-title ee-explore">🌱 Exploration — Spend a few slots learning</div>
+          <div class="ee-desc">Reserve limited surface area for calculated bets: adjacent genres, less-exposed titles, new launches, or representation changes that teach the system something new about the member.</div>
         </div>
       </div>
     </div>