Data analyst · Banking & financial services

Sharad Babar.

Making sense of data.
Making a difference with it.

4 years turning card, payment, lending, and fraud data into dashboards, experiments, and reporting that risk, finance, and operations teams trust, at Citigroup and Capgemini.

Try it · Fraud threshold explorer

Every alert threshold
is a trade-off.

Drag the line. Catch more fraud, or send fewer false alarms to good customers?

Low riskHigh risk
LegitimateFraudFlagged ▸
92%fraud caught
184false alarms / 10k
33%alert precision

Illustrative model. At Citi, an A/B test of a tighter threshold cut false positives by 14% with no rise in fraud losses.

Scroll to discoverSQL · Python · Power BI · TableauPortfolio / 2026

01A little about me

Curiosity meets
business impact.

I'm Sharad, a Data Analyst turning banking data into decisions people can act on.

With 4 years in banking and financial services, I analyze card, payment, lending, deposit, customer, fraud, and collections data, combining SQL, Python, statistics, and a clear understanding of the business behind the numbers.

At Citigroup, I build card risk and collections dashboards, automate MIS reporting, and run data quality checks on regulatory feeds. Previously at Capgemini, I delivered loan and deposit reporting, ETL pipelines, and reconciliations for a global retail banking client.

I own the question end to end: reconciling the data, testing the change or forecast, and presenting the result to risk, finance, and operations teams.

Let's talk data
How I work

From a good question
to a better decision.

01

Understand

Frame the problem and define what success means.

02

Build & validate

Reconcile the data. Test the change. Challenge the assumptions.

03

Make it useful

Publish governed dashboards and communicate the result clearly.

MS
M.S. Computer Science

State University of New York · 2025

4

Years in data analytics

Banking & financial services

14%

Fewer false-positive alerts

Fraud threshold A/B test at Citi

~35%

Fewer open data issues

Regulatory report reconciliation

My focus
Fraud & card risk analyticsRegulatory reporting & data qualityExperimentationForecasting

02Experience

Real problems.
Measurable progress.

Building dashboards, experiments, and reconciled reporting that risk, finance, and operations teams rely on in banking.

Citigroup

Current

Data Analyst

Feb 2026 - Present

New York, USA

Card risk & regulatory analytics

14%fewer false-positive alerts
15+dashboards built
~35%fewer open data issues
  • Built 15+ Tableau and Power BI dashboards for card risk, collections and operations leaders, modeling a star schema with DAX measures, LOD expressions and row-level security; the self-service views retired ~30 recurring email reports.
  • Designed an A/B test with the fraud strategy team to size a tighter transaction-monitoring threshold before rollout, using power analysis and a two-proportion z-test; the chosen rule cut false-positive alerts by 14% with no rise in fraud losses.
  • Reconciled monthly balance and transaction totals between source systems and the reporting warehouse for 20+ regulatory and management reports, running root cause analysis on breaks and reducing open data issues by ~35% within two quarters.
More about this role
  • Automated weekly MIS and management reporting packs with Python (pandas, NumPy) and scheduled SQL jobs, choosing Python over Excel VBA because refresh logic needed version control in Git; prep time fell from two days to under 3 hours.
  • Analyzed quarterly fraud alert and chargeback trends across disputed card transactions with PySpark on the Hadoop data lake, applying hypothesis testing to separate real pattern shifts from seasonal noise and flagging two merchant categories for rule review.
  • Implemented automated data quality checks in Python and SQL (completeness, duplicates, ledger tie-outs) on regulatory reporting feeds, documenting data lineage and metadata in Confluence and tracking remediation in Jira so auditors could trace each field to source.
  • Forecasted monthly collections call volumes with SARIMA time-series models in statsmodels, picked over a moving-average baseline because volumes carried strong month-end seasonality; error held near 8% MAPE, feeding agent staffing plans.

Capgemini

Data Analyst

Jun 2020 - Jul 2023

Bengaluru, India

Retail banking reporting & ETL

1M+customer accounts
~40%faster nightly batch
6%lift in policy renewals
  • Delivered SQL Server and Oracle reporting for a global retail banking client's loan and deposit portfolios (1M+ customer accounts), writing stored procedures and views that fed 25+ recurring operational and compliance reports.
  • Engineered ETL pipelines in Azure Data Factory and SSIS to load core banking extracts into a SQL Server data warehouse; switching to incremental loads cut nightly batch runtime by ~40% and freed the early-morning reporting window.
  • Streamlined month-end reconciliation between general ledger and sub-ledger extracts with Advanced Excel (pivot tables, XLOOKUP, Power Query) and VBA macros, reducing manual matching effort for the finance operations team from three days to one.
More about this role
  • Modeled savings-account churn with logistic regression in scikit-learn, chosen over random forest because relationship managers needed explainable drivers; the top-decile risk list captured about 3x the baseline churn rate.
  • Tested email versus SMS renewal reminders for an insurance client with t-tests and chi-square analysis in SciPy, recommending the channel mix that lifted policy renewals by 6% and became the default for later renewal cycles.
  • Mapped source-to-target data flows and co-wrote business requirement and data mapping documents with client analysts, then built UAT test cases within Agile sprints to validate report logic before each production release.
  • Standardized KPI definitions and a shared data dictionary across client reporting teams under a data governance program, and optimized slow-running SQL queries through indexing, execution plan review and rewriting correlated subqueries as joins.

03The toolkit

A versatile stack.
A focused approach.

The tools and methods behind my work, from raw source extracts to the reports and dashboards leaders rely on.

At the core
SQLPythonPySparkTableauPower BISnowflake
01

Programming & Databases

PythonPandasNumPyAdvanced SQLCTEs & Window FunctionsQuery TuningT-SQLPL/SQLRSASC++UNIX/Linux ShellGit
02

Statistics & Experimentation

Hypothesis TestingA/B Testing DesignPower AnalysisTwo-Proportion Z-TestsT-TestsChi-Square AnalysisTrend AnalysisCohort AnalysisRFM Segmentation
03

Analytics & BI

TableauPower BIDAXPower QueryStar Schema ModelingLookerQlikSenseSSRSAdvanced ExcelVBA MacrosPower Pivot
04

Modeling & Forecasting

scikit-learnSciPystatsmodelsLogistic RegressionRandom ForestSARIMA ForecastingPredictive AnalyticsQ-LearningDeep Q-NetworksMulti-Agent RL
05

Cloud, Big Data & ETL

AWS RedshiftAWS S3AWS EC2SnowflakeBigQueryDatabricksPySparkSpark SQLHadoopHiveAzure Data FactorySSISInformaticaAirflowAlteryx
06

Banking, Risk & Governance

AML/KYC MonitoringCard Risk & ChargebacksCollectionsCCAR / DFASTBasel III/IVSOXDodd-FrankGL ReconciliationData Lineage (Collibra)BRD & Data MappingUAT Design

04Selected projects

The thinking.
The work. The impact.

Two featured banking analytics projects, plus more of my work on GitHub, from SQL analysis and ETL pipelines to machine learning.

From alerts to insight01
01
Ingest & model

AWS S3 → Snowflake star schema

02
Engineer fraud signals

Velocity · Distance · Merchant risk

03
Compare thresholds

Precision vs. recall · Power BI

Fraud analytics · Business intelligence

Card Transaction Fraud Monitoring & Alert Analytics

Fraud teams reviewed rule alerts without one view of where losses concentrate or how thresholds trade catch rate against customer friction. This project builds that view and compares rule thresholds head to head.

1.8M+Card transactions modeled
~20%More fraud caught at same alert volume
Explore the approach
  • Loaded 1.8M+ synthetic card transactions from AWS S3 into Snowflake and modeled a star schema (transaction fact; customer, merchant and date dimensions), orchestrated with Apache Airflow for daily incremental refreshes.
  • Engineered fraud features in SQL and Python (transaction velocity, distance from home, merchant category risk) and compared rule thresholds by precision and recall; a combined velocity-plus-amount rule caught ~20% more fraud at the same alert volume.
  • Published a Power BI dashboard with DAX measures for fraud rate, loss by merchant category, alert precision and hour-of-day heatmaps, with drill-through from the portfolio view down to single transactions.
SnowflakeAWS S3Apache AirflowSQLPythonStar SchemaFeature EngineeringPrecision & RecallPower BIDAX
From onboarding to retention02
01
Cohort retention

PySpark · Spark SQL · RFM

02
Explain churn

Logistic regression · statsmodels

03
Test a retention offer

Fee-waiver A/B · Tableau story

Customer analytics · Experimentation

Digital Banking Churn & Cohort Retention Analysis

A retail bank saw new deposit customers go inactive after onboarding but could not tell when they dropped off or which segments to target. This project traces retention by cohort and tests an offer to win them back.

1M+Account events processed
~2xChurn without early direct deposit
Explore the approach
  • Processed 1M+ simulated monthly account events on Databricks with PySpark, joined them to a 10K-customer public bank churn dataset, and ran cohort analysis of monthly retention curves and RFM segments in Spark SQL.
  • Fitted a logistic regression churn model in statsmodels, finding that customers without direct deposit in the first 60 days churned at roughly 2x the rate of others.
  • Simulated an A/B test of a fee-waiver retention offer with power analysis and a two-proportion z-test, then presented lift, confidence intervals and a targeting recommendation in a Tableau story for business stakeholders.
DatabricksPySparkSpark SQLPythonstatsmodelsCohort AnalysisRFM SegmentationLogistic RegressionA/B TestingTableau

+More on GitHub

Analytics, data engineering, and ML projects.

View all repositories

Retail analytics · BI

Customer Shopping Behavior Analysis

Cleaned and explored 3,900 retail purchases across 18 variables in pandas, answered 10 business questions in PostgreSQL, and built a Power BI dashboard.

Found that male customers drive ~68% of revenue and that subscribers spend no more per purchase than non-subscribers.

PythonPandasPostgreSQLSQLPower BI

Data engineering · ETL

Spotify ETL Pipeline on AWS & Snowflake

Extracts playlist and audio-feature data from the Spotify API into S3, transforms it with AWS Lambda, and loads it into Snowflake through Snowpipe.

Scheduled with EventBridge cron and event-based triggers for fully automated ingestion.

PythonSpotipyAWS S3AWS LambdaEventBridgeSnowflake

SQL analytics · Data modeling

H-1B Visa Applications Analysis (FY2016)

Modeled 50,000+ H-1B petitions in a relational schema with an ER diagram and used SQL to analyze case outcomes, top filing employers, approval rates, and wages by job title.

Shows which employers and roles take the largest share of the annual 85,000-visa cap.

SQLAzure Data StudioER ModelingPythonJupyter

Machine learning · Web app

Student Performance Prediction System

Compared logistic regression, random forest, SVM, decision tree, AdaBoost, and XGBoost to classify final grades into five bands, then served the model in a Flask app.

XGBoost reached ~88.5% validation accuracy and was retrained on the 12 most important features.

Pythonscikit-learnXGBoostFlaskSeaborn

Deep learning · NLP

Emotion Prediction from Text

Classifies text into six emotions with a bidirectional GRU, chosen after comparing RNN, LSTM, and GRU models, served through a FastAPI backend and web interface.

Returns the predicted emotion with a confidence score for every label.

PythonTensorFlow/KerasBiGRUFastAPINLP

Full-stack development

Food Delivery Web App

A MERN food ordering platform with login, cart, Stripe checkout, payment verification, order tracking, and an admin dashboard for menu and orders.

JWT-secured Express API on MongoDB Atlas, with separate customer and admin React apps.

ReactNode.jsExpressMongoDBStripeJWT

05Let's connect

Good questions.
Great possibilities.

Have a data challenge, an interesting role, or an idea worth exploring? I'd love to hear about it.

sharad.b@cvmailhub.com

Open to data analyst opportunities

Start a conversation

Opens a draft in your email app. Review and send it from there.