Skip to content

Selected Work

A selection of data, AI, statistical and analytical systems I have designed and built.

Each project has a full case study covering the problem, the data, the modelling approach, validation, explainability and the decision it supports.

01 — Featured

Predictive Churn Risk Intelligence System

Retail Banking · Machine Learning · Risk Analytics

A full-stack churn prediction system for retail banking that moves from raw customer data through an automated ETL pipeline to explainable risk scores, AI-generated retention strategies and a supervisory report.

Engineered features
18
Engineered features
Models benchmarked
4
Models benchmarked
Evaluation metrics
5
Evaluation metrics
Explainability
SHAP
Explainability
  • Python
  • Scikit-learn
  • SHAP
  • Gemini API
  • ETL Pipeline Design
  • Financial Risk Analytics
  1. 01Customer Data
  2. 02ETL
  3. 0318 Engineered Features
  4. 04Model Benchmarking
  5. 05Best Model
  6. 06SHAP Explainability
  7. 07Risk Segmentation
  8. 08Gemini Recommendations
  9. 09Decision Support
System architecture — illustrative diagram of the churn intelligence pipeline.
02 — Featured

CricAnalytics

Sports Intelligence · Statistical Modelling · Predictive Simulation

A statistical match-prediction platform that runs 5,000 parallel Monte Carlo trials per prediction over 1,000+ historical IPL matches, combining ELO ratings, hypothesis testing and EMA-based season forecasting in a React interface.

Monte Carlo trials
5,000
Monte Carlo trials
Historical matches
1,000+
Historical matches
IPL seasons
19
IPL seasons
Significance test
Two-tailed Z
Significance test
  • Python
  • React 18
  • Monte Carlo Simulation
  • Z-Test
  • ELO Rating
  • Recharts
  • Tailwind CSS
SIMULATED OUTCOME5,000 TRIALS
Illustrative distribution shape — not output from a specific match run.
03 — FeaturedGold Medal · 1st Rank

Defect Reduction & Operational Efficiency

Manufacturing · Lean Six Sigma DMAIC · Quality Analytics

A Lean Six Sigma DMAIC study on the Medium Casing transformer line at Atlanta Electricals Pvt. Ltd. Real defect records were cleaned in pandas/NumPy, ranked with Pareto and 6M root-cause analysis, prioritised with FMEA, and converted into SOPs, Poka-Yoke controls and a 5S programme — cutting the defect rate from 8% to 2% and monthly cost of poor quality from ₹76,000 to ₹22,000.

Defect rate
8% → 2%
Defect rate
Monthly COPQ
₹76K → ₹22K
Monthly COPQ
Cycle time
−40%
Cycle time
Recognition
Gold Medal
Recognition
  • Lean Six Sigma
  • DMAIC
  • Python
  • Pandas
  • NumPy
  • Pareto Analysis
  • FMEA
  • SPC
  • 5S
4126221795WORKMANSHIPDESIGNMATERIALASSEMBLYTESTINGOTHERS120 DEFECTS RECORDEDTOP 2 = 56% CUMULATIVE
Pareto ranking of the 120 recorded defect events at Atlanta Electricals — workmanship and design issues alone account for 56% of all defects.

Also built

Machine Learning · Classification · NASA Open Dataset

Hazardous Asteroid Detection Model

A binary classification model on NASA's near-Earth object dataset that prioritizes recall, because in hazard detection a missed positive costs far more than a false alarm.

  • Python
  • Pandas
  • Scikit-learn
  • Logistic Regression
  • Matplotlib

Portfolio Brief

Looking for the fastest way to review my work?

A three-page executive overview: positioning, proof metrics, flagship work, how I approach problems, toolkit, education and recognition. Readable in the browser, printable to PDF, no email required.

Get My Portfolio Brief