I am a data scientist with robust experience in the banking industry, specializing in customer analytics, predictive modeling, and AI-driven business solutions. My expertise spans customer segmentation, churn prediction, cross-sell analytics, security monitoring, and automation using advanced statistical models, machine learning, and intelligent agents. I have hands-on experience leveraging the OpenAI SDK to build AI-powered tools that extend traditional analytics and streamline data-driven decision-making.
1. Login Pattern Analytics
Project Focus: Predictive Security Analytics using Machine Learning
Challenge: Detecting anomalous login patterns to flag potential password sharing, a critical security issue in banking.
Technical Highlights:
Developed Python pipelines combining rule-based exception logic with Random Forest models.
Engineered features on very large, detailed log data to enhance results.
Complexity & Impact:
Managed imbalanced datasets and optimized hyperparameters for minimal false positives.
Built data-driven exception reports, empowering Internal Audit and IT security teams to intervene quickly.
2. FM Sentiment Analysis
Project Focus: Staff Satisfaction Modeling via Transformer-based NLP
Challenge: Quantifying staff satisfaction from Facility Management (FM) ticket narratives, where feedback was unstructured and subtle.
Technical Highlights:
Deployed Large Language Models (LLMs) like Gemma and BERT to extract nuanced sentiment signals from ticket text.
Complexity & Impact:
Created dashboards to visualize sentiment trends, allowing management to pinpoint service bottlenecks.
Business Value:
Enabled data-backed, targeted improvements in facility services.
3. Agentic Executive Summary Generation (POC)
Project Focus: Automated Reporting with LLM-driven Summarization
Challenge: Synthesizing lengthy audit reports into concise, actionable executive summaries.
Technical Highlights:
Built an agentic solution leveraging LLMs to extract critical audit findings and recommendations.
Calibrated summarization models to preserve regulatory language and context fidelity.
Complexity & Impact:
Addressed the challenge of tailoring summary tone and scope to varied end-users (board vs. audit teams).
Enabled rapid dissemination of key findings to non-technical stakeholders.
Business Value:
Accelerated audit cycles, increased report consumption, and improved compliance follow-up rates.
4. Video Analytics for Gold Loan Evaluation
Project Focus: Computer Vision for Customer Identity Validation (POC)
Challenge: Ensuring compliance by validating customer presence during gold appraisal stage.
Technical Highlights:
Utilized YOLO object detection models on real-time CCTV data for facial recognition and occupancy analysis.
Designed a workflow to match customer images from transaction records with live feed data.
Complexity & Impact:
Overcame image quality, angle variation, and real-time processing constraints.
Business Value:
Improved risk mitigation in high-value loan processes through robust automation.
5. Server Capacity Forecasting with Facebook Prophet
Project Focus: Predictive Analytics for IT Infrastructure
Challenge: Validating the sufficiency of server provisioning ahead of critical deployments.
Technical Highlights:
Applied Facebook Prophet for robust, time series-based server load forecasting.
Integrated operational data feeds and anomaly detection to flag potential capacity issues in advance.
Complexity & Impact:
Dealt with seasonality, irregular access patterns, and data quality issues in infrastructure logs.
Provided scenario-based forecasts for multiple hardware clusters.
Business Value:
Supported proactive resource planning, reducing downtime and improving IT cost efficiency.
Customer Churn Prediction Model:
Project Focus: Customer Retention Analytics using Machine Learning
Feature Engineering:
Crafted advanced behavioral and transactional features—such as withdrawal frequency, declined transactions, online banking engagement, and customer support interactions—to improve prediction accuracy.
Model Development:
Built machine learning models (Random Forest, XGBoost) to detect early signals of churn. Tackled imbalanced data using techniques like SMOTE.
Complexity & Impact:
Generated actionable, data-driven exception reports highlighting high-risk customer segments. These were shared with the Relationship Management and Customer Service teams for timely follow-up.
1. Customer Segmentation Model:
Project Focus: Targeted Customer Understanding Using Segmentation.
Technical Highlights:
Developed Python pipelines using various segmentation techniques.
Impact & Business Value:
Successfully categorized retail customers of the bank into four categories. This helped the bank understand the segment that is suitable for each product to cross-sell.
2. Housing Loan Cross-Sell Model:
Project Focus: Predictive Cross-Sell Analytics using Machine Learning
Technical Highlights:
Selected and engineered relevant customer features predictive of housing loan propensity.
Impact & Business Value:
Overcame class imbalance and ensured high-precision targeting to maximize loan conversions.
Automated extraction and scoring pipelines streamlined sales outreach.
Junior Associate of Indian Institute of Bankers (JAIIB)