Projects | Robert Jean Pierre

Fraud Detection System

6.3M transactions

Built ML pipeline achieving 99.9% fraud recall and 0.9999 ROC-AUC using advanced feature engineering and SMOTE balancing.

Python LightGBM SHAP

0 % Recall

0 ROC-AUC

Technical Deep Dive

Implemented a sophisticated fraud detection pipeline processing 6.3 million financial transactions in real-time. The system uses ensemble methods combining LightGBM with custom feature engineering.

Key Achievements

Reduced false positives by 67% while maintaining 99.9% recall
Implemented SMOTE for handling class imbalance
Deployed SHAP for model interpretability
Processing latency under 100ms per transaction

Healthcare Management System

500+ users

Full-stack system with 20+ normalized tables, role-based access control, and automated drug interaction checking for concurrent users.

PHP MySQL JavaScript

0 DB Tables

0 Active Users

System Architecture

Designed and implemented a comprehensive healthcare management platform serving 500+ concurrent users with role-based access control and real-time drug interaction checking.

Technical Highlights

20+ normalized database tables with optimized indexes
Real-time drug interaction checking algorithm
Role-based access control with 5 permission levels
HIPAA-compliant data encryption

Swin Transformer Research

96.35% accuracy

Comprehensive evaluation of hierarchical vision transformers, demonstrating superiority over CNNs and ViT models.

PyTorch Computer Vision

0 % Accuracy

Image Matching Challenge 2025

CVPR Workshop

Custom pipeline for clustering scenes and estimating camera poses from unstructured image collections.

Python OpenCV PyTorch

CVPR Conference

Calorie Burn Prediction

0.0593 RMSLE

Ensemble model using weighted LightGBM-XGBoost-CatBoost blend with metabolic feature engineering.

Python Ensemble

AI-Powered Game Engine

90%+ win rate

Strategic game AI with deduction logic and pathfinding algorithms, achieving 90%+ win rate.

Python Game AI

Featured Projects