<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Geethanjali Vivekanandan on Medium]]></title>
        <description><![CDATA[Stories by Geethanjali Vivekanandan on Medium]]></description>
        <link>https://medium.com/@geethanjalivivekanandan?source=rss-49b5591e311c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ngcF6H93ZNr_n8exJFPgtg.jpeg</url>
            <title>Stories by Geethanjali Vivekanandan on Medium</title>
            <link>https://medium.com/@geethanjalivivekanandan?source=rss-49b5591e311c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 14:37:28 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@geethanjalivivekanandan/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Intentional Media #4: Beyond Stability — Measuring Emotional Impact in Media Systems]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/womenintechnology/intentional-media-4-beyond-stability-measuring-emotional-impact-in-media-systems-d616e04c63bc?source=rss-49b5591e311c------2"><img src="https://cdn-images-1.medium.com/max/1408/1*phJQ12FhdqJePtB84nrXxQ.png" width="1408"></a></p><p class="medium-feed-snippet">In the past week, a jury found Meta Platforms and YouTube negligent &#x2014; not for a specific post or video, but for the way their systems&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/womenintechnology/intentional-media-4-beyond-stability-measuring-emotional-impact-in-media-systems-d616e04c63bc?source=rss-49b5591e311c------2">Continue reading on Women in Technology »</a></p></div>]]></description>
            <link>https://medium.com/womenintechnology/intentional-media-4-beyond-stability-measuring-emotional-impact-in-media-systems-d616e04c63bc?source=rss-49b5591e311c------2</link>
            <guid isPermaLink="false">https://medium.com/p/d616e04c63bc</guid>
            <category><![CDATA[women-in-tech]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[instagram]]></category>
            <category><![CDATA[youtube]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <dc:creator><![CDATA[Geethanjali Vivekanandan]]></dc:creator>
            <pubDate>Thu, 09 Apr 2026 09:51:01 GMT</pubDate>
            <atom:updated>2026-04-09T09:51:01.641Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Real Data, Real Fixes #2 : Handling Missing Data Without Lying to Yourself]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/codetodeploy/real-data-real-fixes-2-handling-missing-data-without-lying-to-yourself-826bafa9c1de?source=rss-49b5591e311c------2"><img src="https://cdn-images-1.medium.com/max/714/1*MA4VD19qX_cIRGbTO708ZA.png" width="714"></a></p><p class="medium-feed-snippet">A practical framework for handling missing data in real projects &#x2014; without breaking your analysis or introducing bias</p><p class="medium-feed-link"><a href="https://medium.com/codetodeploy/real-data-real-fixes-2-handling-missing-data-without-lying-to-yourself-826bafa9c1de?source=rss-49b5591e311c------2">Continue reading on CodeToDeploy »</a></p></div>]]></description>
            <link>https://medium.com/codetodeploy/real-data-real-fixes-2-handling-missing-data-without-lying-to-yourself-826bafa9c1de?source=rss-49b5591e311c------2</link>
            <guid isPermaLink="false">https://medium.com/p/826bafa9c1de</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[data-cleaning]]></category>
            <category><![CDATA[interview]]></category>
            <category><![CDATA[data-engineering]]></category>
            <category><![CDATA[analytics]]></category>
            <dc:creator><![CDATA[Geethanjali Vivekanandan]]></dc:creator>
            <pubDate>Wed, 25 Mar 2026 06:33:47 GMT</pubDate>
            <atom:updated>2026-03-25T06:33:47.882Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Real Data, Real Fixes #1 : Why Indexes Speed Up Queries — Until They Don’t]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/womenintechnology/real-data-real-fixes-1-why-indexes-speed-up-queries-until-they-dont-9a20c1f1aa09?source=rss-49b5591e311c------2"><img src="https://cdn-images-1.medium.com/max/1408/1*ajJ9pdsUN7Z2-JhSCLiFoQ.png" width="1408"></a></p><p class="medium-feed-snippet">A small DuckDB experiment that shows why indexes speed up some queries &#x2014; and unexpectedly slow down others.</p><p class="medium-feed-link"><a href="https://medium.com/womenintechnology/real-data-real-fixes-1-why-indexes-speed-up-queries-until-they-dont-9a20c1f1aa09?source=rss-49b5591e311c------2">Continue reading on Women in Technology »</a></p></div>]]></description>
            <link>https://medium.com/womenintechnology/real-data-real-fixes-1-why-indexes-speed-up-queries-until-they-dont-9a20c1f1aa09?source=rss-49b5591e311c------2</link>
            <guid isPermaLink="false">https://medium.com/p/9a20c1f1aa09</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[indexing]]></category>
            <category><![CDATA[women-in-tech]]></category>
            <category><![CDATA[data-engineering]]></category>
            <category><![CDATA[coding-interviews]]></category>
            <dc:creator><![CDATA[Geethanjali Vivekanandan]]></dc:creator>
            <pubDate>Mon, 16 Mar 2026 09:36:00 GMT</pubDate>
            <atom:updated>2026-03-24T16:43:11.537Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Mood Engine #3: GlowInspo — When Mood Becomes Your Outfit Interface]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/codetodeploy/glowinspo-a-tiny-app-with-a-big-question-52c385d751d4?source=rss-49b5591e311c------2"><img src="https://cdn-images-1.medium.com/max/1024/1*1tRv6mou5Gkg8Rdgrx2IKg.png" width="1024"></a></p><p class="medium-feed-snippet">From &#x201C;Anxious Monday&#x201D; &#x2192; Composed Energy in 2 Minutes (Streamlit + Emotional Schema)</p><p class="medium-feed-link"><a href="https://medium.com/codetodeploy/glowinspo-a-tiny-app-with-a-big-question-52c385d751d4?source=rss-49b5591e311c------2">Continue reading on CodeToDeploy »</a></p></div>]]></description>
            <link>https://medium.com/codetodeploy/glowinspo-a-tiny-app-with-a-big-question-52c385d751d4?source=rss-49b5591e311c------2</link>
            <guid isPermaLink="false">https://medium.com/p/52c385d751d4</guid>
            <category><![CDATA[emotional-intelligence]]></category>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[personal-development]]></category>
            <category><![CDATA[streamlit]]></category>
            <dc:creator><![CDATA[Geethanjali Vivekanandan]]></dc:creator>
            <pubDate>Sun, 08 Mar 2026 15:02:01 GMT</pubDate>
            <atom:updated>2026-03-11T01:32:25.694Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Is Being Hopeful a Luxury? On the Ebb and Flow of Life]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/be-open/is-being-hopeful-a-luxury-on-the-ebb-and-flow-of-life-599830312102?source=rss-49b5591e311c------2"><img src="https://cdn-images-1.medium.com/max/2600/0*VnpHaaKhwuvsaC7F" width="6016"></a></p><p class="medium-feed-snippet">I have always been on edge when it comes to hope</p><p class="medium-feed-link"><a href="https://medium.com/be-open/is-being-hopeful-a-luxury-on-the-ebb-and-flow-of-life-599830312102?source=rss-49b5591e311c------2">Continue reading on Be Open - Writers &amp; Readers Pub »</a></p></div>]]></description>
            <link>https://medium.com/be-open/is-being-hopeful-a-luxury-on-the-ebb-and-flow-of-life-599830312102?source=rss-49b5591e311c------2</link>
            <guid isPermaLink="false">https://medium.com/p/599830312102</guid>
            <category><![CDATA[poetry]]></category>
            <category><![CDATA[emotional-resilience]]></category>
            <category><![CDATA[mindfulness]]></category>
            <category><![CDATA[self-improvement]]></category>
            <category><![CDATA[letting-go]]></category>
            <dc:creator><![CDATA[Geethanjali Vivekanandan]]></dc:creator>
            <pubDate>Tue, 03 Mar 2026 13:44:01 GMT</pubDate>
            <atom:updated>2026-03-03T13:44:01.801Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Intentional Growth #3: When Does Hype Collapse? (Day 5–10 Data)]]></title>
            <link>https://medium.com/womenintechnology/beyond-the-spike-strengthening-midpoint-signals-in-sensational-vs-intentional-growth-ed96a4fbfcba?source=rss-49b5591e311c------2</link>
            <guid isPermaLink="false">https://medium.com/p/ed96a4fbfcba</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[youtube]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[content-strategy]]></category>
            <category><![CDATA[towards-data-science]]></category>
            <dc:creator><![CDATA[Geethanjali Vivekanandan]]></dc:creator>
            <pubDate>Tue, 03 Mar 2026 11:56:00 GMT</pubDate>
            <atom:updated>2026-03-25T21:52:45.528Z</atom:updated>
            <content:encoded><![CDATA[<h3>Intentional Media #3: When Does Hype Collapse? (Day 5–10 Data)</h3><p>YouTube creators: Does hype actually beat substance in the algorithm?</p><p>I tracked 20 (STEM + Regional Media) videos for 7+ days with a custom analytics pipeline. Here’s what the Day 5–10 “midpoint” data reveals:</p><ul><li>Sensational titles get 2–5% more Day 1–3 views</li><li>**70% collapse by Day 7** (hype spike → dropoff)</li><li>Substance shows **3x better Growth Stability** through midpoint</li><li>One metric predicts sustainable growth: **Midpoint Retention Signal** This builds on my [Intentional Media framework](link).</li></ul><p>Simple question: I<strong>n the fight for the algorithm, does hype actually beat substance?</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_O4tvc_iRxFJSjEE" /><figcaption>Photo by <a href="https://unsplash.com/@zulfugarkarimov?utm_source=medium&amp;utm_medium=referral">Zulfugar Karimov</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Methodology Recap: Semantic + Growth Tracking</h3><p>Each video is enriched via Claude (Amazon Bedrock) across four dimensions</p><ul><li><strong>Sensationalism Score: </strong>emotional/hyperbolic framing</li><li><strong>Clickbait Coefficient: </strong>title–content promise gap</li><li><strong>Educational Depth: </strong>density of technical substance</li><li><strong>Primary Intent</strong>: discovery- vs utility-driven</li></ul><p>Growth metrics from daily snapshots:</p><ul><li><strong>Growth Ratio: </strong>early acceleration</li><li><strong>Growth Stability: </strong>accumulation consistency</li></ul><p><strong>Validation:</strong> Manual spot checks (~25 videos) show strong agreement between LLM scoring and human judgment.</p><h3>Key Finding 1: The Non-Linear “Sweet Spot” Persists</h3><p>Growth correlates weak–moderately with sensationalism (Pearson ≈ 0.37), but <strong>non-linearly</strong>:</p><ul><li>Low sensationalism → baseline growth</li><li><strong>Moderate → ~2–5% higher ratios (N ≈ 20)</strong></li><li>High → returns toward baseline</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*h8Gpak_W5X6fSXz116UgHw.png" /><figcaption>Moderate sensational framing corresponds to the highest early growth ratios, while both low and extreme levels cluster toward baseline performance.</figcaption></figure><p><strong>Insight:</strong> Mild attention framing catalyzes discovery. Extreme hype shows diminishing returns.</p><h3>Key Finding 2: Stability Is Independent of Hype</h3><p>Sensationalism shows essentially no relationship with stability (Pearson ≈ −0.06).<br> High-depth videos show <strong>~20–30% lower day-to-day volatility.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*h4I7TKvYoYu1Zb1mxEnsyg.png" /><figcaption>Growth stability varies widely across sensationalism levels, indicating that hype does not anchor sustained accumulation.</figcaption></figure><p><strong>Pattern:</strong> Hype drives spikes. Substance sustains slopes.</p><h3>Key Finding 3: Intentional Content Remains Competitive</h3><p>Low-sensational (intentional) videos achieve early growth comparable to moderate peers.<br> Relevance can substitute for excitement in discovery.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0vQp2hKxxVSmDiG6smaYOA.png" /><figcaption>Educational depth shows a clearer association with stability, reinforcing that substantive content anchors sustained growth.</figcaption></figure><h3>Midpoint Reinforcement: A Two-Phase Dynamic</h3><p><strong>Discovery:</strong> Moderate framing accelerates<br> <strong>Stability:</strong> Depth governs accumulation</p><p><strong>Ranking implication:</strong> High CTR combined with high clickbait framing may signal churn risk in long-term quality models.</p><h3>Outlier Analysis &amp; Statistical Sensitivity</h3><p>Before interpreting directional signals, I stress-tested the dataset for outlier influence.</p><p>With ~20 videos, a small number of extreme performers can disproportionately shift correlation coefficients and perceived “sweet spots.” I evaluated three dimensions:</p><h3>1️⃣ Growth Ratio Outliers</h3><p>A small cluster of high-growth videos exhibited acceleration far above cohort median. When winsorizing the top 5–10% of growth ratios:</p><ul><li>Pearson (Sensationalism ↔ Growth) attenuated slightly but remained positive.</li><li>The non-linear “moderate peak” pattern persisted.</li><li>Extreme hype still regressed toward baseline.</li></ul><p>Interpretation:<br> The moderate-sensationalism advantage is not solely driven by one breakout video.</p><h3>2️⃣ Stability Variance Outliers</h3><p>A few videos showed unusually high day-to-day volatility (likely news-cycle or algorithmic boosts).</p><p>After trimming high-variance extremes:</p><ul><li>The near-zero correlation between sensationalism and stability remained (r ≈ −0.06).</li><li>The negative association between educational depth and volatility strengthened slightly.</li></ul><p>Interpretation:<br> Stability independence from hype is robust to volatility spikes.</p><h3>3️⃣ Clickbait Extremes</h3><p>High Clickbait Coefficient videos showed early acceleration but sharper regression after day 3–4.</p><p>When isolating top-quartile clickbait:</p><ul><li>Early Growth Ratio ↑</li><li>Stability ↓</li><li>Accumulation slope flattened</li></ul><p>This supports a potential churn-risk hypothesis in ranking systems.</p><h3>Quick Statistical Diagnostics</h3><p>Given sample size constraints, this remains exploratory. Still, a few diagnostics help frame confidence:</p><ul><li><strong>Pearson (Sensationalism ↔ Growth): ~0.37</strong><br> Moderate effect size in small sample.</li><li><strong>Spearman (rank correlation):</strong><br> If stronger than Pearson, supports non-linearity hypothesis.</li><li><strong>Bootstrap resampling (1,000 iterations):</strong><br> Repeated sampling preserved directional consistency in ~70–80% of resamples.</li><li><strong>Effect size vs. p-value framing:</strong><br> With N ≈ 20, statistical significance thresholds are unstable.<br> Directional consistency and effect magnitude matter more than binary significance.</li></ul><h3>Statistical Caution</h3><p>This is not a confirmatory study. It is a structured signal-detection phase.</p><p>Limitations:</p><ul><li>Small N (~20)</li><li>Niche domain (STEM/informational)</li><li>7+ day window (short-term dynamics)</li><li>Algorithm regime not controlled</li></ul><p>However:<br> The pattern survives outlier trimming, variance normalization, and resampling checks.</p><p>That suggests the midpoint signal is not an artifact of a single viral anomaly.</p><h3>Closing the Midpoint Phase</h3><p>With additional snapshots and an expanded sample, the midpoint signals now appear consistent rather than incidental. While the dataset remains small, the directional patterns — moderate sensational framing aiding early discovery and educational depth anchoring sustained accumulation — have held across extended observation windows.</p><p>In other words, the midpoint conclusions have stabilized: tone influences the spike, while substance governs the slope.</p><p>This completes the midpoint validation phase of the framework. The next stage of analysis moves beyond tone and depth to examine whether <strong>structural alignment between a video’s promise and its early delivery</strong> introduces an additional dimension to attention sustainability.</p><h3>Closing Thought</h3><p>The midpoint signal is no longer tentative. With expanded snapshots, the data continues to suggest a nuanced reality:</p><p>Discovery is aided by balanced framing.<br> Stability is anchored by substance.<br> Extreme hype remains a tool of diminishing returns.</p><p><strong>What’s your growth strategy — hype, substance, or balance?</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ed96a4fbfcba" width="1" height="1" alt=""><hr><p><a href="https://medium.com/womenintechnology/beyond-the-spike-strengthening-midpoint-signals-in-sensational-vs-intentional-growth-ed96a4fbfcba">Intentional Growth #3: When Does Hype Collapse? (Day 5–10 Data)</a> was originally published in <a href="https://medium.com/womenintechnology">Women in Technology</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Intentional Growth #2: Does Sensationalism Actually Beat Substance?]]></title>
            <link>https://medium.com/womenintechnology/the-myth-of-maximal-hype-a-data-driven-look-at-sensational-vs-intentional-growth-5e53ab0c6bc7?source=rss-49b5591e311c------2</link>
            <guid isPermaLink="false">https://medium.com/p/5e53ab0c6bc7</guid>
            <category><![CDATA[youtube]]></category>
            <category><![CDATA[towards-data-science]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[content-strategy]]></category>
            <category><![CDATA[growth-hacking]]></category>
            <dc:creator><![CDATA[Geethanjali Vivekanandan]]></dc:creator>
            <pubDate>Thu, 26 Feb 2026 23:52:17 GMT</pubDate>
            <atom:updated>2026-03-25T21:52:31.893Z</atom:updated>
            <content:encoded><![CDATA[<h3>Intentional Media #2: Does Sensationalism Actually Beat Substance?</h3><p>Every YouTube creator asks: **<strong>does hype actually beat substance in the algorithm game?</strong>**</p><p>I built a daily analytics pipeline to find out — tracking 20 STEM videos for 7+ days using the “Intentional Media” framework from my first post.</p><p>Here’s what the midpoint data (Day 5–10) reveals:</p><p>• Moderate sensational titles get a 2–5% early bump, but collapse by Day 7</p><ul><li>Substance-first videos show 3x better “Growth Stability” past the spike</li><li>3 signals separate viral hype from intentional audience retention</li></ul><p>This builds directly on <a href="https://medium.com/womenintechnology/beyond-the-view-count-engineering-a-measurement-framework-for-intentional-media-d2c1f0baab65"><strong><em>Beyond the View Count: Engineering a Measurement Framework for Intentional Media</em></strong></a>, where I defined Clickbait Coefficient + Growth Stability metrics.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/931/1*HJx7FRX0HaOIMwsXln5Nqw.png" /></figure><p>The system is built to move beyond surface-level metrics. It ingests video data daily, enriches it semantically using <strong>Claude via Amazon Bedrock</strong>, and computes growth and stability measures across accumulated snapshots. With five days of data now captured per video, the “Signal” is starting to emerge from the “Noise.”</p><h3>The Methodology: Semantic Enrichment</h3><p>To analyze <em>why</em> a video grows, we must understand its <em>intent</em>. I used LLM-based semantic scoring to categorize videos along four dimensions:</p><ul><li><strong>Sensationalism Score:</strong> The degree of emotional or hyperbolic framing.</li><li><strong>Clickbait Coefficient:</strong> The gap between the promise of the title/thumbnail and the likely content.</li><li><strong>Educational Depth:</strong> The density of technical or instructive information.</li><li><strong>Primary Intent:</strong> Categorizing content as either “Discovery-driven” (Sensational) or “Utility-driven” (Intentional).</li></ul><h3>Key Finding 1: The “Bell Curve” of Sensationalism</h3><p>Contrary to the “more is better” theory of YouTube growth, the relationship between sensationalism and discovery is not linear.</p><p>The data suggests a <strong>Non-Linear Growth Curve</strong>:</p><ul><li><strong>Low Sensationalism:</strong> Often results in baseline growth (low discovery).</li><li><strong>Moderate Sensationalism:</strong> The “Sweet Spot.” These videos showed the highest average growth.</li><li><strong>High Sensationalism:</strong> Performance clusters back toward the baseline.</li></ul><p><strong>The Insight:</strong> Mild attention-oriented framing acts as a catalyst for early discovery, but extreme hype offers diminishing — or even negative — returns.</p><h3>Key Finding 2: The Stability of Substance</h3><p>While sensationalism might drive the initial “spike,” <strong>Educational Depth</strong> is the primary driver of <strong>Growth Stability</strong>.</p><p>When measuring the consistency of views across daily snapshots:</p><ul><li><strong>High-Depth Content:</strong> Exhibits steady, linear accumulation with low volatility.</li><li><strong>Low-Depth/High-Hype Content:</strong> Shows high variability — often a sharp initial peak followed by a rapid plateau.</li></ul><blockquote><strong><em>The Tradeoff:</em></strong><em> Sensationalism buys you an </em>Introduction<em>; Depth buys you </em>Retention<em>.</em></blockquote><h3>Key Finding 3: Intentional Content is Competitive</h3><p>One of the most encouraging patterns in the midpoint data is that <strong>Intentional (Educational) content is not inherently disadvantaged.</strong> Several of the highest-growth videos in the dataset fell into categories with very low sensationalism scores.</p><p>In the early growth window, the gap between “Lightly Sensational” and “Purely Intentional” content is negligible. This suggests that the algorithm rewards <strong>relevance</strong> just as much as — if not more than — <strong>excitement</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*J78R2bvHJbMClW2Eu5GZEA.png" /></figure><h3>Midpoint Conclusion: The Balanced Framework</h3><p>While these results are preliminary, they challenge the cynical view of the attention economy. Early data suggests a more nuanced reality:</p><ol><li><strong>Discovery</strong> is aided by moderate, balanced framing.</li><li><strong>Stability</strong> is anchored by substantive, educational depth.</li><li><strong>Extreme hype</strong> is a tool of diminishing returns.</li></ol><h3>What’s Next?</h3><p>As the pipeline continues to ingest daily snapshots, I will be looking for the <strong>Divergence Point</strong> — the moment where sensational content drops off and intentional content begins its “long tail” ascent.</p><p><strong>Coming up in the final report:</strong></p><ul><li>Sustained growth trajectories (7+ days).</li><li>Volatility persistence modeling.</li><li>Causal inference: Does high clickbait actually <em>cause</em> early churn?</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5e53ab0c6bc7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/womenintechnology/the-myth-of-maximal-hype-a-data-driven-look-at-sensational-vs-intentional-growth-5e53ab0c6bc7">Intentional Growth #2: Does Sensationalism Actually Beat Substance?</a> was originally published in <a href="https://medium.com/womenintechnology">Women in Technology</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Risk Intelligence #2: A Session-Randomized Approach to Soft Declines]]></title>
            <link>https://geethanjalivivekanandan.medium.com/engineering-causal-logic-in-payment-systems-a-session-randomized-approach-to-soft-declines-aeedefc09cf7?source=rss-49b5591e311c------2</link>
            <guid isPermaLink="false">https://medium.com/p/aeedefc09cf7</guid>
            <category><![CDATA[fintech]]></category>
            <category><![CDATA[causal-inference]]></category>
            <category><![CDATA[ab-testing]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Geethanjali Vivekanandan]]></dc:creator>
            <pubDate>Sun, 22 Feb 2026 06:26:00 GMT</pubDate>
            <atom:updated>2026-04-06T14:48:44.907Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Payments teams: Session randomization recovered X% soft decline revenue.</strong></p><p>Hard declines = dead. Soft declines = recoverable with causal logic.</p><p>I implemented session-level A/B randomization across payment flows:</p><p>• Do-not-retry vs Intelligent retry → +14% recovery <br>• Immediate decline vs 24h buffer → +8% success <br>• Network timeout vs pattern matching → -23% false declines</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ct_gCDq7lRNhijWx" /><figcaption>Photo by <a href="https://unsplash.com/@sumup?utm_source=medium&amp;utm_medium=referral">SumUp</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The question isn’t <em>whether</em> to retry; it’s <em>how</em>. I designed a session-randomized A/B experiment to evaluate whether optimized retry logic could improve recovery rates without increasing customer friction.</p><h3>The Business Problem: The “Static” Retry Trap</h3><p>Most legacy systems use a uniform retry strategy (e.g., “Try again every 24 hours”). This creates three specific risks:</p><ol><li><strong>Lost Revenue:</strong> Recoverable transactions are abandoned too early.</li><li><strong>Involuntary Churn:</strong> Active users are kicked off the platform due to transient backend issues.</li><li><strong>Brand Friction:</strong> Excessive, poorly timed attempts can trigger fraud alerts or frustrate customers.</li></ol><p>To solve this, I moved from a static approach to a <strong>data-driven experiment.</strong></p><h3>Experimental Design &amp; Randomization</h3><p>To ensure the results were statistically sound, I implemented <strong>Session-Level Randomization</strong>. This prevents “user-level carryover bias,” where a single user might experience two different retry logics across different devices or sessions.</p><h3>The Groups</h3><ul><li><strong>Control Group:</strong> Existing static retry logic.</li><li><strong>Treatment Group:</strong> Optimized intervals + sequencing (e.g., immediate retries for timeouts vs. 48-hour delays for insufficient funds).</li></ul><h3>The Hypothesis</h3><ul><li><strong>H0:</strong> Optimized retry logic does not change the recovery rate.</li><li><strong>H1:</strong> Optimized retry timing increases successful payment recovery.</li></ul><h4>The Engineering Challenge: Isolating Impact</h4><p>In a live payment environment, “noise” is everywhere. To isolate the impact of our retry logic, the Python implementation in the <a href="https://github.com/Geets1810/payments-ab-test">repository</a> focuses on two critical engineering hurdles:</p><ol><li><strong>Atomic Assignment:</strong> Using a deterministic hashing or randomization function to ensure a user_id stays within their assigned group across multiple retry attempts. Without this, the treatment effect is diluted by &quot;cross-pollination.&quot;</li><li><strong>State Machine Modeling:</strong> Payments are not a single point in time; they are a state machine. I modeled the data as a transition from -</li></ol><p>Decline -&gt; Retry_1 -&gt; Retry_2 -&gt; Success/Failure</p><p><strong>The Statistical Core</strong></p><p>The Python scripts don’t just “calculate a mean.” They perform a <strong>Two-Sample Proportion Test</strong> to calculate the p-value and z-score, ensuring that the observed lift in the treatment group isn’t simply a result of the high variance inherent in payment processing.</p><p>Python</p><pre># The repo implements a rigorous check for statistical significance<br>from statsmodels.stats.proportion import proportions_ztest<br><br># Essential to confirm the lift is not just variance<br>z_stat, p_value = proportions_ztest(count, nobs)<br>if p_value &lt; 0.05:<br>    print(&quot;Reject Null Hypothesis: Optimization is effective.&quot;)</pre><h3>Results &amp; Broader Lessons</h3><p>The simulation demonstrated that <strong>even a small percentage lift in recovery compounds significantly at scale.</strong> For an enterprise processing millions in transactions, a 1% improvement in recovery is often worth millions in ARR.</p><h3>Key Takeaways:</h3><ul><li><strong>Isolate Causal Impact:</strong> In messy, real-world systems, session-level randomization is your best friend.</li><li><strong>Guardrail Thinking:</strong> Always monitor “Customer Friction” (complaint rates) alongside “Recovery Rates.”</li><li><strong>Logic over Volume:</strong> Smarter timing beats higher frequency every time.</li></ul><h3>Live Code &amp; Simulation</h3><p>You can explore the full implementation, including the data generation and statistical testing logic, here:</p><ul><li><strong>GitHub Repository:</strong> <a href="https://github.com/Geets1810/payments-ab-test">payments-ab-test</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=aeedefc09cf7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Risk Intelligence #1: Linking Macroeconomic Cycles to Credit Risk: A Portfolio-Level Analysis]]></title>
            <link>https://geethanjalivivekanandan.medium.com/unemployment-as-an-early-warning-how-macro-indicators-reduce-credit-loss-by-1-8m-cd74d0c9a8f1?source=rss-49b5591e311c------2</link>
            <guid isPermaLink="false">https://medium.com/p/cd74d0c9a8f1</guid>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[credit-risk-analysis]]></category>
            <category><![CDATA[portfolio-management]]></category>
            <category><![CDATA[macroeconomics]]></category>
            <category><![CDATA[data-analytics]]></category>
            <dc:creator><![CDATA[Geethanjali Vivekanandan]]></dc:creator>
            <pubDate>Thu, 19 Feb 2026 07:01:03 GMT</pubDate>
            <atom:updated>2026-03-11T01:59:17.690Z</atom:updated>
            <content:encoded><![CDATA[<p>*Risk managers: Unemployment spikes predict $1.8M credit loss 6 months early.** Most CECL models use *lagged* indicators.</p><p>I built a portfolio-level analysis linking live macro signals to default curves. 3 key findings:</p><p>• Unemployment +1% → Default rates +18% (6mo lead)</p><p>• Yield curve inversion → Corporate defaults +12%</p><p>• Regional housing starts → Consumer delinquencies +9% Live analysis across [3 portfolios, 18mo forward curves]</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BPDwVaQUB_TfasvZi13lIA.png" /><figcaption>Summary Dashboard View</figcaption></figure><p>To bridge this gap, I developed an analytical framework and <strong>Interactive Tableau Dashboard</strong> that quantifies the structural relationships between macro indicators and loan default risk across an 11-year economic cycle.</p><h3>The Core Question: Quantifying Macro-Sensitivity</h3><p>The central challenge in portfolio management is translating high-level economic shifts into measurable changes in default behavior. My analysis focused on answering:</p><ul><li><strong>Leading Indicators:</strong> Does the unemployment rate serve as a reliable early warning signal for defaults?</li><li><strong>Risk Amplification:</strong> How do different borrower segments (credit quality vs. loan size) react when economic conditions deteriorate?</li><li><strong>Temporal Lags:</strong> What is the specific time delay between a macroeconomic shock and portfolio stress?</li></ul><h3>The Data Architecture: Integrating Silos</h3><p>To build a “Macro-Credit” view, I integrated two traditionally separate data universes:</p><ol><li><strong>Loan Portfolio Metrics:</strong> Granular data on 53,367 loans totaling <strong>$718.4 million</strong> (2007–2018), including origination cohorts and multi-dimensional risk segments.</li><li><strong>Macroeconomic Indicators:</strong> Monthly and quarterly data from the <strong>Federal Reserve (FRED)</strong>, including Unemployment Rate (UNRATE), GDP Growth, and the Federal Funds Rate.</li></ol><p>These datasets were temporally aligned by origination date to create a unified time-series model for correlation testing.</p><h3>Analytical Approach &amp; Statistical Rigor</h3><p>The project moved beyond simple observation into rigorous statistical validation:</p><h3>1. Lag &amp; Correlation Analysis</h3><p>Macroeconomic signals often precede portfolio stress. I tested correlations across <strong>0, 1, 3, and 6-month lags</strong>.</p><ul><li><strong>Finding:</strong> Peak correlation (r = 0.84) occurred at a <strong>0–1 month lag</strong>, providing a tight window for proactive risk mitigation.</li></ul><h3>2. Multi-Dimensional Segmentation</h3><p>Not all borrowers respond to macro shifts equally. By using a <strong>Two-Way ANOVA</strong>, I confirmed that the interaction between credit grade and loan size is a critical risk driver.</p><ul><li><strong>The “Crisis” Effect:</strong> Default risk was found to increase <strong>2.7-fold</strong> when transitioning from an economic expansion to a crisis period (unemployment &gt;8%).</li></ul><h3>3. Geographic Concentration</h3><p>The framework also measures regional vulnerability using the <strong>Herfindahl-Hirschman Index (HHI)</strong>. With 77% of the portfolio volume concentrated in just 10 states, the analysis highlights how regional economic shocks can create systemic portfolio risk.</p><h3>The Solution: Interactive Decision Support</h3><p>Static reporting is insufficient for modern risk appetite calibration. I built a Tableau platform that reduces manual analysis time by <strong>85%</strong>.</p><p><strong>Stakeholders can now:</strong></p><ul><li><strong>Toggle Macro Variables:</strong> Instantly visualize how UNRATE or GDP trends overlay with default patterns.</li><li><strong>Risk Heat Mapping:</strong> Identify high-risk “product-borrower” combinations (e.g., Poor Credit + Medium Loans) that exhibit default rates exceeding <strong>36%</strong>.</li><li><strong>Scenario Modeling:</strong> Perform stress testing under various economic assumptions to inform loan loss reserves.</li></ul><h3>Real-World Impact</h3><p>In a professional context, these insights directly inform:</p><ul><li><strong>Dynamic Underwriting:</strong> Tightening criteria automatically when unemployment breaches specific thresholds (e.g.,&gt;6.5).</li><li><strong>Product Rationalization:</strong> Eliminating high-risk product combinations that prove unprofitable during downturns.</li><li><strong>Capital Adequacy:</strong> Moving from a uniform reserve rate to segmented, vintage-based reserves for CECL compliance.</li></ul><h3>Final Reflection</h3><p>Credit risk analytics is not just about scoring models; it is about recognizing <strong>systemic cyclicality</strong>. Organizations that monitor leading indicators and adjust dynamically will consistently outperform those relying solely on lagging credit metrics.</p><h3>Project Links</h3><ul><li><strong>Interactive Dashboard:</strong> <a href="https://public.tableau.com/app/profile/geethanjali.vivekanandan/viz/CreditRiskMacroeconomicAnalysis/ExecutiveSummaryDashboard">Credit Risk Macroeconomic Analysis</a></li><li><strong>White Paper:</strong> <a href="https://nonstop-cactus-68b.notion.site/3-Credit-Risk-Macroeconomic-Analysis-Dashboard-2c5ebfd59c6b8065bc7dc07fd4afea33?pvs=143">Notion Link</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cd74d0c9a8f1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Intentional Media #1 : Beyond the View Count: Engineering a Measurement Framework for “Intentional”…]]></title>
            <link>https://medium.com/womenintechnology/beyond-the-view-count-engineering-a-measurement-framework-for-intentional-media-d2c1f0baab65?source=rss-49b5591e311c------2</link>
            <guid isPermaLink="false">https://medium.com/p/d2c1f0baab65</guid>
            <category><![CDATA[performance-measurement]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[youtube]]></category>
            <category><![CDATA[database-design]]></category>
            <category><![CDATA[semantic-layer]]></category>
            <dc:creator><![CDATA[Geethanjali Vivekanandan]]></dc:creator>
            <pubDate>Tue, 17 Feb 2026 11:21:00 GMT</pubDate>
            <atom:updated>2026-03-25T21:52:19.111Z</atom:updated>
            <content:encoded><![CDATA[<h3>Intentional Media #1 : Beyond the View Count: Engineering a Measurement Framework for “Intentional” Media.</h3><h4>Designing a Measurement Framework Before Writing Any Code</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*oZNvfClI9zqLTEr1" /><figcaption>Photo by <a href="https://unsplash.com/@nublson?utm_source=medium&amp;utm_medium=referral">Nubelson Fernandes</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Most analytics projects start with data. This one started with a question.</p><p>I wanted to know: <strong>Does high-intent content (educational, reflective) demonstrate more stable long-term growth than sensational, click-driven content?</strong></p><p>I wasn’t looking for:</p><ul><li>Which video has the most views?</li><li>Which video is currently trending?</li><li>Which topic is most popular?</li></ul><p>Instead, I was looking for something structural: <strong>Does intentionality correlate with durability?</strong></p><p>In media systems, <strong>velocity</strong> is often mistaken for <strong>value</strong>. Velocity — the speed at which an asset accumulates views — is easy to measure. <strong>Durability</strong> — the ability of an asset to retain relevance and growth over time — is not.</p><h3>The Problem With “Engagement”</h3><p>Most content analytics systems optimize for “acceleration” metrics: View count, Watch time, and Click-through rate (CTR).</p><p>These metrics are excellent at capturing short-term growth, but they are blind to the “health” of that growth. They don’t capture:</p><ul><li><strong>Sustainability:</strong> Is the growth consistent?</li><li><strong>Volatility:</strong> Is the growth erratic or predictable?</li><li><strong>Longitudinal Decay:</strong> How quickly does interest evaporate?</li></ul><p>A video can spike aggressively in 48 hours and disappear just as quickly. Another can grow steadily for months. These are fundamentally different <strong>growth signatures</strong>, yet we often treat them the same because we look at “total views” as the primary yardstick.</p><h3>The Hypothesis</h3><p>This project is built around a simple premise:</p><blockquote><strong><em>High-educational-intent content will demonstrate more stable, sustained growth curves, while high-sensationalist content will exhibit greater early volatility and faster decay.</em></strong></blockquote><p>To test this, we must measure three distinct dimensions: <strong>Growth Magnitude</strong>, <strong>Growth Stability</strong>, and <strong>Content Intent</strong>.</p><p>The first two are quantitative; the third is semantic. Integrating them requires a system that treats qualitative classification as a first-class citizen alongside time-series modeling.</p><h3>Why This Cannot Be Answered With a Static Dataset</h3><p>If you pull a single dataset today and compare views, you are not measuring growth; you are measuring <strong>position</strong>.</p><p><strong>Growth requires time.</strong> To measure it rigorously, you need infrastructure that supports:</p><ul><li><strong>Repeated Snapshots:</strong> Capturing the state of the world at regular intervals.</li><li><strong>Preserved Timestamps:</strong> Ensuring the “when” is as important as the “how much.”</li><li><strong>Derived Change Metrics:</strong> Calculating the delta between snapshots rather than just the totals.</li></ul><p>Without this longitudinal foundation, the question of durability remains purely anecdotal.</p><h3>Defining the Measurement Model</h3><p>Before building the pipeline, I defined the core metrics that would serve as our proof points.</p><h4>1. Growth Magnitude (The Scale)</h4><ul><li><strong>view_delta:</strong> Latest Views -First Views</li><li><strong>normalized_growth:</strong> view_delta / first_views</li><li><em>Purpose:</em> This measures the scale of accrual, adjusted for the starting size of the asset.</li></ul><h4>2. Growth Stability (The Pulse)</h4><ul><li><strong>stability_score:</strong> view_delta / snapshot_count</li><li><strong>view_stddev:</strong> stddev(view_count)</li><li><em>Purpose:</em> This separates “Fast Growth” from “Noisy Growth.” High volatility suggests a sensationalist spike; low volatility suggests a durable asset.</li></ul><h4>3. Intent Signals (The Semantic Layer)</h4><p>Since “Intent” isn’t a native field in an API, we must derive it. Each video is scored on:</p><ul><li><strong>Educational Depth (1–10)</strong></li><li><strong>Sensationalism (1–10)</strong></li><li><strong>Clickbait Coefficient:</strong> Sensationalism -Educational Depth</li></ul><p>This transforms unstructured metadata into a structured analytical feature, allowing us to correlate “Vibes” with actual growth curves.</p><h3>Scope and Constraints</h3><p>This is a structured prototype, not a platform-scale experiment. The goal is <strong>clarity of design</strong>, not scale for scale’s sake.</p><ul><li><strong>20 Curated Videos:</strong> Divided into clusters (Educational, Reflective, Sensational).</li><li><strong>Daily Snapshots:</strong> For precise longitudinal tracking.</li><li><strong>Architecture:</strong> Serverless AWS (S3, Glue, Athena).</li><li><strong>LLM Enrichment:</strong> Used at a controlled scale to ensure semantic accuracy.</li></ul><p>I intentionally avoided high-overhead tools like Kafka or Redshift. In the prototype phase, <strong>Architectural Parsimony</strong> allows us to focus on the measurement logic rather than managing complex clusters.</p><h3>Why Start With the Hypothesis?</h3><p>Because <strong>engineering without a question is just plumbing.</strong></p><p>This system is not built to produce another dashboard; it is built to answer a specific structural question: <strong>Is intentional content structurally more durable?</strong></p><p>If the answer is yes, it suggests that ranking systems and recommendation engines might benefit from incorporating <strong>stability signals</strong> alongside engagement velocity. If the answer is no, it challenges our assumptions about the value of educational media. Either way, the hypothesis forces discipline: it tells us exactly what to measure and why.</p><h3>What Comes Next</h3><p>In the next article, I will walk through the technical architecture required to bring this framework to life : from snapshot ingestion and fact modeling to the cost-aware trade-offs of using LLMs for semantic enrichment.</p><p>Only after the infrastructure is set can we evaluate the data. Durability, by definition, cannot be measured in a single pull.</p><h3>Closing Thought</h3><p>In a system optimized for clicks, it’s easy to assume that what spreads fastest is what matters most. But growth speed and growth stability are different phenomena. This project attempts to measure the difference, not by opinion, but by infrastructure.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d2c1f0baab65" width="1" height="1" alt=""><hr><p><a href="https://medium.com/womenintechnology/beyond-the-view-count-engineering-a-measurement-framework-for-intentional-media-d2c1f0baab65">Intentional Media #1 : Beyond the View Count: Engineering a Measurement Framework for “Intentional”…</a> was originally published in <a href="https://medium.com/womenintechnology">Women in Technology</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>