<?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 Chandrashekhar Buradkar on Medium]]></title>
        <description><![CDATA[Stories by Chandrashekhar Buradkar on Medium]]></description>
        <link>https://medium.com/@buradkar.lokesh?source=rss-203b3a36cace------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*KluhqzqZRBUF4qFgXFLcQQ.jpeg</url>
            <title>Stories by Chandrashekhar Buradkar on Medium</title>
            <link>https://medium.com/@buradkar.lokesh?source=rss-203b3a36cace------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 05:49:20 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@buradkar.lokesh/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[Agentic AI Beyond the Demo: Building Agents That Recover From Failure]]></title>
            <link>https://medium.com/@buradkar.lokesh/agentic-ai-beyond-the-demo-building-agents-that-recover-from-failure-b4811c5effd6?source=rss-203b3a36cace------2</link>
            <guid isPermaLink="false">https://medium.com/p/b4811c5effd6</guid>
            <dc:creator><![CDATA[Chandrashekhar Buradkar]]></dc:creator>
            <pubDate>Wed, 28 Oct 2026 07:59:58 GMT</pubDate>
            <atom:updated>2026-10-28T07:59:58.131Z</atom:updated>
            <content:encoded><![CDATA[<p>An AI agent checks a customer’s booking, updates a reservation, and sends a confirmation.</p><p>The demo looks impressive.</p><p>Now imagine the reservation update succeeds, but the confirmation tool times out. Should the agent repeat the entire process? Send another message? Tell the customer the booking failed?</p><p>This is where agentic AI becomes an engineering problem.</p><p>A useful agent needs a recovery plan alongside its action plan.</p><p>What makes agentic AI different?</p><p>An AI agent can choose tools, take actions, inspect results, and adjust its next step toward a goal. That flexibility helps with tasks whose steps cannot all be predicted in advance. It also creates opportunities for errors to compound. Anthropic’s guidance highlights environmental feedback, stopping conditions, and testing as essential parts of agent design. (anthropic.com)</p><p>Consider an agent investigating a decline in hotel bookings. It might query a database, compare properties, check cancellations, and prepare a report.</p><p>What happens if one data source is unavailable – or two sources disagree?</p><p>Here are five design choices I would prioritize.</p><ol><li>Define success before execution</li></ol><p>“Investigate the booking decline” is too vague.</p><p>A better task specifies the period, comparison baseline, required data, and expected output. It also tells the agent how to report missing information.</p><p>Without clear completion criteria, an agent can produce a convincing report while leaving the actual question unanswered.</p><p>2. Record what actually happened</p><p>Track each meaningful action and its outcome.</p><p>In our reservation example, the system should record that the booking update succeeded while message delivery remains uncertain. Recovery can then target the unresolved step.</p><p>Tracing helps developers inspect tool calls, handoffs, and failures. OpenAI’s Agents SDK provides this capability, although applications still need to define their own business success criteria. (openai.github.io)</p><p>3. Make retries safe</p><p>A timeout does not prove an action failed.</p><p>Before repeating a booking change or payment request, check its status. Use a unique operation identifier so the receiving system can recognize duplicate requests.</p><p>Otherwise, a temporary connection problem can become a duplicate transaction.</p><p>4. Set boundaries around actions</p><p>Reading booking data and issuing a refund require different permissions.</p><p>Enforce access limits in the tools and underlying systems. Add approval checkpoints where business policy requires them. Tool-level validation matters because checks on the initial prompt and final response do not cover every intermediate action. (openai.github.io)</p><p>5. Test recovery, not just completion</p><p>Include scenarios such as:</p><ul><li>A tool times out after completing its action.</li><li>* A required data source is unavailable.</li><li>* The agent receives conflicting results.</li><li>* A user changes the request midway through execution.</li></ul><p>Measure successful completion, duplicate actions, recovery time, cost per completed task, and the quality of human handoffs.</p><p>Build for the interrupted workflow</p><p>For your first agent, choose one narrow business process. Give it limited tools, explicit success criteria, and a defined recovery path.</p><p>Then deliberately interrupt it.</p><p>If it can identify what succeeded, explain what remains uncertain, and resume safely, you have a stronger foundation for real operational use.</p><p>The most useful agent is one whose work you can verify – and whose failures you can recover from.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b4811c5effd6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Statistics for Banking: Essential Concepts and Python Code for Data-Driven Insights]]></title>
            <link>https://medium.com/@buradkar.lokesh/statistics-for-banking-essential-concepts-and-python-code-for-data-driven-insights-97ae8528cdf2?source=rss-203b3a36cace------2</link>
            <guid isPermaLink="false">https://medium.com/p/97ae8528cdf2</guid>
            <category><![CDATA[python]]></category>
            <category><![CDATA[statistics]]></category>
            <category><![CDATA[banking]]></category>
            <dc:creator><![CDATA[Chandrashekhar Buradkar]]></dc:creator>
            <pubDate>Sun, 03 Nov 2024 11:11:00 GMT</pubDate>
            <atom:updated>2024-11-03T11:11:00.510Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Introduction</strong></p><p>Statistics is fundamental in the banking sector, supporting activities such as risk assessment, customer segmentation, and fraud detection. In this article, we’ll explore key statistical concepts applied in banking, complete with Python code to illustrate how these techniques can turn data into insights. Let’s dive into how data science powers modern banking.</p><ol><li><strong>Descriptive Statistics for Customer Analysis</strong></li></ol><p>Descriptive statistics summarize large volumes of transaction data, giving banks insights into customer behavior, average account balances, and more.</p><p>Example: Calculate basic statistics on customer account balances.</p><p>import pandas as pd</p><p># Sample data: Customer account balances</p><p>data = {‘CustomerID’: [1, 2, 3, 4, 5],</p><p>. ‘AccountBalance’: [5000, 15000, 12000, 20000, 30000]}</p><p>df = pd.DataFrame(data)</p><p># Descriptive statistics</p><p>mean_balance = df[‘AccountBalance’].mean()</p><p>median_balance = df[‘AccountBalance’].median()</p><p>std_dev_balance = df[‘AccountBalance’].std()</p><p>print(f”Mean Balance: {mean_balance}”)</p><p>print(f”Median Balance: {median_balance}”)</p><p>print(f”Standard Deviation: {std_dev_balance}”)</p><p><strong>2. Probability Theory for Credit Risk Analysis</strong></p><p>Probability theory helps banks estimate the likelihood of loan defaults or market shifts, essential for risk assessment.</p><p>Example: Calculate the probability of a customer defaulting based on historical data.</p><p>import numpy as np</p><p># Sample data: 0 = no default, 1 = default</p><p>defaults = [0, 0, 1, 0, 1, 1, 0, 0, 1, 0]</p><p>probability_of_default = np.mean(defaults)</p><p>print(f”Probability of Default: {probability_of_default}”)</p><p><strong>3. Inferential Statistics for Customer Insights</strong></p><p>Inferential statistics allows banks to make generalizations about customer behavior, even with limited survey data.</p><p>Example: Perform a hypothesis test to determine if customer satisfaction scores differ between two bank branches.</p><p>from scipy.stats import ttest_ind</p><p># Sample data: satisfaction scores from two branches</p><p>branch_a = [8, 7, 9, 8, 8, 9, 7, 7]</p><p>branch_b = [6, 5, 6, 5, 7, 6, 6, 5]</p><p># Hypothesis test (two-sample t-test)</p><p>t_stat, p_value = ttest_ind(branch_a, branch_b)</p><p>print(f”T-statistic: {t_stat}, P-value: {p_value}”)</p><p>if p_value &lt; 0.05:</p><p>. print(“Significant difference in satisfaction scores between branches.”)</p><p>else:</p><p>. print(“No significant difference in satisfaction scores.”)</p><p><strong>4. Regression Analysis for Predicting Creditworthiness</strong></p><p>Regression helps banks predict outcomes based on customer attributes, such as income and credit history. This is commonly used in credit scoring.</p><p>Example: A simple linear regression model to predict a customer’s likelihood of default based on account balance.</p><p>from sklearn.linear_model import LinearRegression</p><p>import numpy as np</p><p># Sample data</p><p>balances = np.array([5000, 15000, 12000, 20000, 30000]).reshape(-1, 1)</p><p>defaults = np.array([0, 1, 0, 1, 1]) # 0 = no default, 1 = default</p><p># Fit a linear regression model</p><p>model = LinearRegression()</p><p>model.fit(balances, defaults)</p><p># Predict the probability of default for a new balance</p><p>new_balance = np.array([[25000]])</p><p>predicted_default = model.predict(new_balance)</p><p>print(f”Predicted Default Probability for Balance 25000: {predicted_default[0]:.2f}”)</p><p><strong>5. Time Series Analysis for Interest Rate Forecasting</strong></p><p>Time series analysis helps banks forecast trends, such as interest rates or stock prices, over time. This informs investment and risk strategies.</p><p>Example: Using a moving average to smooth fluctuations in interest rate data.</p><p>import pandas as pd</p><p># Sample data: Monthly interest rates</p><p>data = {‘Month’: pd.date_range(start=’1/1/2023&#39;, periods=12, freq=’M’),</p><p>. ‘InterestRate’: [3.5, 3.6, 3.7, 3.8, 4.0, 4.2, 4.3, 4.4, 4.5, 4.7, 4.8, 5.0]}</p><p>df = pd.DataFrame(data)</p><p>df.set_index(‘Month’, inplace=True)</p><p># Calculate a 3-month moving average</p><p>df[‘MovingAverage’] = df[‘InterestRate’].rolling(window=3).mean()</p><p>print(df)</p><p><strong>6. Anomaly Detection for Fraud Prevention</strong></p><p>Fraud detection involves identifying unusual patterns in transactions. Statistics can help banks detect anomalies that may indicate fraudulent activity.</p><p>Example: Z-score to identify outliers in transaction amounts.</p><p>from scipy.stats import zscore</p><p># Sample transaction amounts</p><p>transactions = [100, 200, 150, 10000, 180, 250, 130, 5000, 170, 160]</p><p>z_scores = zscore(transactions)</p><p># Identify transactions with z-score &gt; 2 or &lt; -2 (potential anomalies)</p><p>outliers = [x for x, z in zip(transactions, z_scores) if abs(z) &gt; 2]</p><p>print(f”Potential Fraudulent Transactions: {outliers}”)</p><p><strong>Conclusion</strong></p><p>Understanding and applying statistical concepts in banking offers a powerful way to enhance decision-making and operational efficiency. From risk management to customer analysis, statistics transforms raw data into actionable insights. By integrating these techniques with Python, professionals in the banking industry can drive data-driven decisions that improve outcomes for both the bank and its customers.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=97ae8528cdf2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Statistics for Banking: Essential Concepts and Applications]]></title>
            <link>https://medium.com/@buradkar.lokesh/statistics-for-banking-essential-concepts-and-applications-a98a88556431?source=rss-203b3a36cace------2</link>
            <guid isPermaLink="false">https://medium.com/p/a98a88556431</guid>
            <category><![CDATA[statistics]]></category>
            <dc:creator><![CDATA[Chandrashekhar Buradkar]]></dc:creator>
            <pubDate>Sun, 03 Nov 2024 11:02:25 GMT</pubDate>
            <atom:updated>2024-11-03T11:03:28.654Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Introduction</strong></p><p>In the banking industry, statistics plays a crucial role in analyzing data for risk assessment, customer insights, fraud detection, and more. This article explores the key statistical concepts and methods relevant to banking, providing an overview of why they’re essential and how they’re applied in the industry.</p><ol><li><strong>Why Statistics is Important in Banking</strong></li></ol><p>•	Risk Management: Banks use statistics to assess the likelihood of loan defaults, credit risks, and market fluctuations.</p><p>•	Customer Analytics: Statistical methods help in segmenting customers, understanding their behavior, and improving service personalization.</p><p>•	Fraud Detection: By identifying patterns and anomalies in transaction data, statistics aids in fraud prevention.</p><p>•	Investment Analysis: Statistics enables banks to model market trends and make informed investment decisions.</p><p><strong>2. Key Concepts in a Statistics Course for Banking</strong></p><p>•	Descriptive Statistics: Summarizing customer transaction data, average balances, spending behavior, and loan performance using metrics like mean, median, standard deviation, and variance.</p><p>•	Probability Theory: Understanding the probability of events such as loan defaults or stock price changes, which forms the basis for credit scoring and risk assessments.</p><p>•	Inferential Statistics: Drawing insights from customer surveys and smaller datasets to make decisions about larger customer groups, essential for market research and trend analysis.</p><p>•	Regression Analysis: Used for predicting creditworthiness, customer retention, and spending habits by analyzing relationships between variables (e.g., income level and credit usage).</p><p>•	Time Series Analysis: Analyzing trends over time, such as forecasting interest rates, stock prices, and economic indicators.</p><p><strong>3. Statistical Methods Used in Banking</strong></p><p>•	Logistic Regression: Commonly used for credit scoring, helping banks predict the likelihood of default based on customer demographics and financial history.</p><p>•	Clustering Analysis: Segmenting customers into groups based on similar characteristics, enabling banks to tailor marketing and product offerings.</p><p>•	Hypothesis Testing: Testing marketing campaign effectiveness, like comparing customer engagement between two promotional offers to see which has a statistically significant impact.</p><p>•	Survival Analysis: Used in loan or mortgage retention analysis, predicting when a customer might close an account or pay off a loan early.</p><p><strong>4. Applications of Statistics in Banking</strong></p><p>•	Credit Scoring and Loan Approval: Banks use statistical models to assess credit risk by analyzing historical data on loan repayment patterns and customer profiles.</p><p>•	Fraud Detection and Prevention: Statistical techniques, like anomaly detection, help banks identify unusual transactions that may indicate fraud.</p><p>•	Market Risk Analysis: Understanding exposure to market fluctuations by modeling asset values and portfolio risks.</p><p>•	Customer Retention: Predictive analytics to understand customer churn rates, enabling banks to take proactive steps to retain valuable clients.</p><p>•	Product Development: Statistical analysis of customer data to create new banking products, such as tailored savings accounts or credit offers.</p><p>5. Tools and Software for Banking Statistics</p><p>•	Statistical Software: R and Python (using libraries like pandas, scikit-learn) for modeling, as well as SAS and SQL for data handling and analysis.</p><p>•	Big Data Platforms: Hadoop, Spark, and similar platforms for handling vast amounts of transaction and customer data.</p><p>•	Business Intelligence Tools: Tools like Tableau and Power BI for visualizing data insights and creating dashboards for decision-making.</p><p><strong>6. Best Practices for Success in Banking Statistics</strong></p><p>•	Data Quality Assurance: Ensuring data accuracy is critical, as financial decisions rely heavily on data integrity.</p><p>•	Continuous Learning: The financial industry and statistical methods evolve quickly; staying updated is essential.</p><p>•	Ethics and Compliance: Given the sensitive nature of financial data, understanding regulatory standards and ethical considerations is key.</p><p>•	Focus on Interpretability: Statistical models must be interpretable for stakeholders to make informed decisions, especially in risk management.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a98a88556431" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Predict customer who is going to default next month]]></title>
            <link>https://medium.com/@buradkar.lokesh/how-to-predict-customer-who-is-going-to-default-next-month-9f60693021df?source=rss-203b3a36cace------2</link>
            <guid isPermaLink="false">https://medium.com/p/9f60693021df</guid>
            <dc:creator><![CDATA[Chandrashekhar Buradkar]]></dc:creator>
            <pubDate>Fri, 23 Aug 2024 15:48:28 GMT</pubDate>
            <atom:updated>2024-08-23T16:16:32.557Z</atom:updated>
            <content:encoded><![CDATA[<h3>How to Predict customer who is going to default in 1 yr after their application get approved</h3><p>Steps</p><ul><li>Prepare data by taking application level data , 1 will be customer who’s loan approved and 0 will be good customer</li><li>Prepare data by taking other features which include customer transactional data , product holding data , demographics data</li><li>Pull data in python</li><li>Preprocess the data by checking null value , by performing EDA.</li><li>Create feature based on based features</li><li>Split the into train, test and OOT validation</li><li>Perform scaling , feature selection on training data,</li><li>Check correlation and remove highly correlated variables</li><li>Handle class imbalanced data using oversampling , understanding,smote methods</li><li>Run logistic regression , random forest , xgboost, neural network models and check the f1 score , precision , recall for train , test</li><li>Tune the model parameter.</li><li>Check decile wise default rate for top deciles , good model able to capture almost all customer in top 3 deciles</li><li>Check same for oot validation data</li></ul><p>How to use above model realtime</p><ul><li>So at the time of application once custome apply for loan then will predict customer propensity score of default</li><li>Next we check in which deciles in fall</li><li>Next if customer fall in top 3 decile will directly reject the customer</li><li>If customer fall in lower decile let’s say in decile 4 to 7 then will increase rate of interest and check all the customer background along with Ccris score</li><li>If customer fall in lower decile then will accept the customer loan application</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9f60693021df" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>