<?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 Srujana Yadav on Medium]]></title>
        <description><![CDATA[Stories by Srujana Yadav on Medium]]></description>
        <link>https://medium.com/@srujanayadav202005?source=rss-347082aebd05------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*O_DKhzxwnjH2OOKVXE1hVA.png</url>
            <title>Stories by Srujana Yadav on Medium</title>
            <link>https://medium.com/@srujanayadav202005?source=rss-347082aebd05------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 03:15:32 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@srujanayadav202005/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[Turning 12 Years of Historical Data into a Data Science Project]]></title>
            <link>https://medium.com/@srujanayadav202005/turning-12-years-of-historical-data-into-a-data-science-project-a4be33ae8802?source=rss-347082aebd05------2</link>
            <guid isPermaLink="false">https://medium.com/p/a4be33ae8802</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[python-programming]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Srujana Yadav]]></dc:creator>
            <pubDate>Tue, 17 Nov 2026 05:00:27 GMT</pubDate>
            <atom:updated>2026-11-17T05:00:27.820Z</atom:updated>
            <content:encoded><![CDATA[<p><em>An account of how I collected real-world data, automated the pipeline, and began building a model to predict daily pilgrim arrivals at one of India’s most visited spiritual destinations.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/855/1*D18Iv4jiaWrBio59TdtQag.png" /><figcaption><em>Tirumala, one of India’s most visited spiritual destinations.</em></figcaption></figure><h3>Introduction</h3><p>Thousands of devotees visit the Tirumala Tirupati Temple every day. Managing crowds of this scale demands careful planning across resource allocation, transportation, and crowd control. This raised a question: <strong>can historical data help anticipate tomorrow’s pilgrim footfall?</strong></p><p>Instead of using a standard sample dataset, I chose to work on a real-world problem with real historical records.</p><h3>1. Problem Statement</h3><p>The objective is to predict the number of pilgrims visiting Tirumala on a given day. Footfall is influenced by several factors:</p><ul><li>Weekdays versus weekends</li><li>Festivals and special occasions</li><li>Brahmotsavam and other major religious events</li><li>Seasonal variation</li><li>Historical arrival patterns</li></ul><p>Understanding these drivers is the first step toward identifying meaningful patterns in pilgrim movement.</p><h3>2. Data Collection</h3><p>Sourcing historical data was one of the most interesting challenges. I used the official TTD news portal (<a href="https://news.tirumala.org/">https://news.tirumala.org/</a>) and developed a Python-based web scraping workflow to extract daily pilgrim statistics from published reports.</p><p>The aim was to build a multi-year dataset rather than rely on a small, pre-existing one. I also automated the collection process so that newly published records can be appended to the dataset without manual effort.</p><h3>3. Data Cleaning and Preprocessing</h3><p>Raw data is rarely model-ready. Key preprocessing steps included:</p><ul><li>Converting date columns to datetime format</li><li>Sorting records chronologically</li><li>Identifying and removing duplicate records</li><li>Checking for missing dates and missing values</li><li>Validating daily pilgrim counts</li><li>Preparing features for analysis and modelling</li></ul><p>At one analysis stage, the dataset contained <strong>3,572 records and 17 columns</strong>.</p><h3>4. Feature Engineering</h3><p>Rather than relying only on past pilgrim counts, I engineered features that capture external influences on daily arrivals:</p><ul><li>Date-based features</li><li>Weekend indicator</li><li>Festival indicator</li><li>Brahmotsavam indicator</li><li>Seven-day rolling average</li><li>Google Trends score</li></ul><p>The goal is for the model to learn both historical arrival patterns and the external factors that shape them.</p><h3>5. Exploratory Data Analysis</h3><p>Before modelling, I examined the dataset to understand its behaviour. Initial findings:</p><p>MetricValueTotal recorded arrivals 243.5 millionAverage daily arrivals 68,172Highest daily count110,662Lowest daily count29,295</p><p>These figures show considerable day-to-day variation. I also compared weekday and weekend patterns and assessed the impact of special occasions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/667/1*uQ0aJouArTMhfBQr5o6yzA.png" /><figcaption><em>Exploring historical pilgrim footfall patterns through data visualisation.</em></figcaption></figure><h3>6. Building the Prediction System</h3><p>The end goal is an interactive system where a user enters a date and receives an estimated pilgrim footfall for it.</p><ul><li><strong>Input:</strong> 25 December 2026</li><li><strong>Output:</strong> Predicted pilgrim arrivals for that date</li></ul><p>This makes the tool more practical than one that only produces forecasts for a fixed set of dates. The pipeline is built on historical data, engineered features, and time-dependent patterns.</p><h3>7. Challenges</h3><p>Real-world projects rarely go smoothly. The main challenges were:</p><ol><li>Consolidating historical information from multiple, differently structured reports</li><li>Handling inconsistent date formats</li><li>Identifying and addressing missing dates</li><li>Quantifying the influence of festivals and special events</li><li>Designing features that are genuinely predictive</li><li>Building a reliable automated data collection workflow</li></ol><p>These taught me that data science involves far more than training a model. Data collection, validation, preprocessing, and a clear understanding of the problem are equally critical.</p><h3>8. Key Learnings</h3><ul><li>Python for data collection and processing</li><li>Web scraping for extracting real-world data</li><li>Pandas for data cleaning and manipulation</li><li>Exploratory data analysis for pattern discovery</li><li>Feature engineering to improve model inputs</li><li>Time-series forecasting concepts</li><li>The importance of data validation and automation</li></ul><p>Above all, I learned how to approach a real-world problem end to end from a data science perspective.</p><h3>Conclusion</h3><p>This is an ongoing project. My aim is not just to predict a number, but to explore how data can support better planning and decision-making in a real operational environment.</p><p>Working with Tirumala footfall data has given me a view of the full data science lifecycle, from raw data collection to a working prediction system. The real value of data science lies in turning raw data into meaningful insights and practical solutions.</p><p><strong>The journey from raw data to real-world predictions has only just begun…..</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a4be33ae8802" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[If Claude Can Analyze My Data in Minutes, Why Do Companies Still Need Data Analysts?]]></title>
            <link>https://medium.com/@srujanayadav202005/if-claude-can-analyze-my-data-in-minutes-why-do-companies-still-need-data-analysts-8db1ccd0ad7a?source=rss-347082aebd05------2</link>
            <guid isPermaLink="false">https://medium.com/p/8db1ccd0ad7a</guid>
            <dc:creator><![CDATA[Srujana Yadav]]></dc:creator>
            <pubDate>Wed, 07 Oct 2026 18:40:01 GMT</pubDate>
            <atom:updated>2026-10-07T18:40:01.641Z</atom:updated>
            <content:encoded><![CDATA[<p>AI can clean, explore, and visualize data faster than ever. But here’s what it still can’t do.</p><p>For years, my work has followed the same cycle:</p><p>Collect → Clean → Explore → Analyze → Visualize → Identify the problem → Recommend a solution.</p><p>At first, it was exciting. But project after project, it began to feel repetitive.</p><p>Then Claude arrived.</p><p>I could hand it a messy dataset and ask it to clean the data, perform exploratory data analysis (EDA), write Python code, build visualizations, run statistical tests, and explain the findings.</p><p>It could do all of this in minutes.</p><p>That raised a question I couldn’t ignore:</p><p>If AI can do so much of a data analyst’s work, why do companies still need data analysts?</p><p>Rather than debate it in theory, let’s look at a real example.</p><p><strong>The Case: A Gym That Started Losing Customers</strong></p><p>Imagine a gym with multiple branches.</p><p>For the first few months, everything went well. Memberships were growing, attendance was strong, and the business was profitable.</p><p>Then the trend reversed. Fewer people were visiting, cancellations increased, and revenue declined.</p><p>The owner came to me with one simple question:</p><p>“What went wrong?”</p><p>The data wasn’t ready for analysis.</p><p>It was spread across multiple Excel files from different branches, with inconsistent formats, mismatched column names, messy dates, and duplicate records.</p><p>So the first step was clear: clean the data.</p><p>I combined the files, standardized the columns, corrected data types, handled missing values, and removed duplicates.</p><p>Then I moved into EDA, examining:</p><p>• Membership trends<br>• Attendance<br>• Cancellations<br>• Revenue<br>• Branch performance<br>• Membership types<br>• Customer behavior</p><p>Once I had a clean, reliable dataset, I gave the same data to Claude.</p><p>What Claude Did</p><p>Claude did a lot, and it did it well.</p><p>It quickly profiled the dataset, identified trends, generated visualizations, and highlighted several possible explanations for the decline, including:</p><p>• Declining attendance<br>• Rising cancellations<br>• Weaker performance in specific branches<br>• Shifts in membership behavior</p><p>It even suggested actions the business could take.</p><p>To be completely honest, Claude was very useful. If I had done all of this technical work manually, it would have cost me hours.</p><p>But there was a catch.</p><p>I had asked, “What is the issue?”</p><p>Claude gave me several possible answers.</p><p>What it found were patterns, not necessarily the root cause.</p><p>And that distinction matters.</p><p>A Pattern Is Not a Cause</p><p>Suppose Claude reports:</p><p>“Gym attendance decreased by 23%.”</p><p>That’s a valuable insight.</p><p>But the real question is: why?</p><p>Maybe:</p><p>• The gym changed its operating hours.<br>• A popular trainer left.<br>• A competitor opened nearby.<br>• Membership prices increased.<br>• The equipment became outdated.<br>• The decline is seasonal.<br>• Attendance dropped because members had already cancelled.</p><p>The dataset can show what happened. It doesn’t necessarily explain why it happened.</p><p>And often, the most important information isn’t in the dataset at all.</p><p>That is where the human analyst becomes essential.</p><p>Claude Sees the Data. The Analyst Sees the Business.</p><p>For me, this is the biggest difference.</p><p>Claude can tell me:</p><p>“Branch A’s revenue declined by 18%.”</p><p>But I can go to the owner and ask:</p><p>“What changed in Branch A three months ago?”</p><p>And the answer might be:</p><p>“Our best trainer left.”</p><p>Suddenly, the entire analysis changes.</p><p>And that context may never have appeared in a spreadsheet.</p><p>A good analyst talks to stakeholders, understands how the business operates, questions assumptions, and connects the numbers to reality.</p><p>That’s something that can’t be solved simply by running more code.</p><p>So, Can Claude Replace a Data Analyst?</p><p>I don’t think the answer is a simple yes or no.</p><p>AI can replace, or dramatically reduce, much of the repetitive work:</p><p>• Data cleaning<br>• Coding<br>• EDA<br>• Visualization<br>• Statistical calculations<br>• Documentation</p><p>And honestly, that’s a positive shift.</p><p>Why should analysts spend hours on routine tasks if AI can handle them?</p><p>The real value of an analyst is moving toward higher-order work:</p><p>• Asking the right questions<br>• Understanding the business context<br>• Validating results<br>• Identifying the true cause<br>• Challenging assumptions<br>• Recommending what the business should do next</p><p>Claude can help me analyze the data, but I still have to judge whether the analysis makes sense.</p><p>That’s why I don’t see Claude as my replacement.</p><p>I see it as my co-analyst.</p><p>The Takeaway</p><p><strong>The future isn’t necessarily:</strong></p><p>Claude vs. Data Analyst</p><p><strong>It is more likely:</strong></p><p>Data Analyst + Claude</p><p>And perhaps the real competition won’t be between humans and AI.</p><p>It will be between analysts who know how to use AI and analysts who don’t.</p><p>AI isn’t necessarily taking the analyst out of the loop. It may simply be changing what the analyst is expected to do.</p><p>Maybe that’s not something we should be afraid of.</p><p>Maybe it’s an opportunity to spend less time cleaning spreadsheets and more time thinking about the problems those spreadsheets are trying to tell us about.</p><p>What’s your take? Is AI a threat or a teammate for data professionals? Share your thoughts in the comments.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8db1ccd0ad7a" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>