Case studies

A portfolio of projects, each told in a different business context.

Each case below focuses on the problem, the solution, and the outcome.

Selected portfolio

Project context, challenge, approach, and outcome.

Ready to review

Selected work

Case studies spanning insurance, finance, manufacturing, and AI product delivery.

Each card is structured to make the context obvious and the business relevance easy to scan.

Insurance AI

Policy adviser prediction system

Built to identify advisers likely to stay active by analyzing behavior, collaboration patterns, and engagement signals.

Challenge

Retain high-value advisers before churn becomes visible to leadership.

Approach

Feature engineering, time-series thinking, Logistic Regression, K-Means, and an Azure-hosted Flask application with Power BI integration.

Outcome

Enabled targeted engagement actions and a more proactive retention workflow.

PythonPandasNumPyScikit-LearnFlaskDockerAzurePower BI

Insurance Operations

Adjuster policy-selling intelligence

Designed a machine learning model to predict when an adjuster would likely sell an insurance policy product to a client.

Challenge

Surface sales opportunities from operational data without slowing field workflows.

Approach

Modeling with feature selection, validation, and deployment-friendly data pipelines connected to enterprise tooling.

Outcome

Improved decision support for field teams and sales prioritization.

PythonSQLAzure Data PipelineDocker

Generative AI

Multi-agent domain assistant

Created a routed AI assistant that answers questions across Wikipedia, finance, medical, and insurance domains.

Challenge

Provide accurate, context-aware responses through multiple specialized agents instead of one generic model.

Approach

LangChain, LangGraph, prompt-based routing, retrieval support, and modular orchestration.

Outcome

Delivered a flexible assistant architecture that can scale to more domains.

LangChainLangGraphGroqPythonAPIs

Insurance AI

RAG-based insurance chatbot

Built an intelligent customer-facing chatbot for policy selection, claims guidance, and premium-related queries.

Challenge

Make insurance information instantly accessible while keeping responses domain-aware and relevant.

Approach

RAG pipelines, embeddings, fine-tuned LLM workflows, and document-backed retrieval with low-latency serving.

Outcome

Improved response quality and customer experience with a more useful self-service layer.

OpenAILLaMA 2GemmaFAISSChromaDBFlaskAzure

Computer Vision

Damage detection and segmentation POC

Developed a deep learning proof of concept for damage estimation, classification, and segmentation in images.

Challenge

Reduce manual inspection effort in claims, automotive, and property assessment workflows.

Approach

YOLOv8-based detection, segmentation techniques, training optimization, and API-ready deployment design.

Outcome

Reduced inspection time and created a stronger foundation for automated claims evaluation.

YOLOv8TensorFlowOpenCVFlaskDockerAzure

Finance Analytics

Customer retention propensity model

Built a model to estimate whether a newly onboarded user would actively trade during the first month after signup.

Challenge

Spot customers at risk of inactivity and improve early engagement.

Approach

Behavioral analysis, logistic modeling, random forest methods, and Power BI reporting for sales actioning.

Outcome

Helped teams focus outreach on higher-value customers with better conversion potential.

PythonSQLPower BIRandom ForestLogistic Regression

Manufacturing Forecasting

Inventory requirement forecasting

Created a planning model to forecast inventory needs using stock, production, and supply chain data.

Challenge

Balance dead stock risk against shortage risk in a production environment.

Approach

Historical analysis, demand forecasting logic, and a Flask web interface deployed in the cloud.

Outcome

Improved procurement planning and stock visibility for operations teams.

PythonFlaskDockerAzureSQL

Retail Banking

Fixed deposit propensity model

Developed a classification model to identify customers most likely to open a high-value fixed deposit product.

Challenge

Help banking teams prioritize outreach using behavioral and financial indicators.

Approach

Machine learning classification, customer segmentation, and an Azure-deployed application layer.

Outcome

Improved targeting for personalized offers and product conversion planning.

PythonFlaskDockerAzure App ServiceSQL

Sports Analytics

Performance intelligence dashboard

Built a Power BI reporting experience combining official match, player, and team performance data.

Challenge

Turn raw sports statistics into a visually clear analysis experience for decision-makers.

Approach

Data collection, metric modeling, bar charts, heatmaps, and radar-style skill comparisons.

Outcome

Created a better view of trends, strengths, and head-to-head patterns.

Power BIData ModelingVisualizationAnalytics