Spanning two high-stakes domains - financial intelligence and clinical AI - each project reflects an end-to-end ML mindset: from raw data and model design to deployed, measurable real-world impact.
[ 01 ] Finance
🛡️
Financial Risk & Fraud Detection System
End-to-end fraud intelligence platform trained on 10M+ transactions - combining supervised classification, anomaly detection, and a RAG-enabled GenAI layer for real-time financial analysis. Achieved 92% accuracy, +12% fraud recall, and a 15% reduction in loan-approval bias through Responsible AI techniques.
PyTorchXGBoostAzure MLRAGLLMsDockerSQLTableau
↗
[ 02 ] Healthcare
🏥
Clinical Chart Review Assistant (RAG + LLM)
Production RAG pipeline using FLAN-T5, Phi-3, and ClinicalBERT (PEFT/LoRA) to auto-generate SOAP note summaries and flag medication safety issues. Reduced hallucination rate by 40% through retrieval grounding - delivering a reliable, audit-ready clinical intelligence layer at scale.
FLAN-T5ClinicalBERTLangChainChromaDBSageMakerFastAPIStreamlit
↗
[ 03 ] Finance
📉
Customer Churn Prediction & Personalization Engine
Churn-prediction system built on 500K+ customer records using advanced feature engineering and ensemble methods. Delivered personalization-driven behavioral insights that reduced modeled churn risk by 12%, paired with interactive Power BI dashboards for real-time stakeholder decision-making.
Scikit-learnXGBoostPandasSQLPower BITableau
↗
[ 04 ] Healthcare
📋
GenAI Clinical Summarization & Compliance Engine
PPO-based RL pipeline with toxicity-aware rewards enforcing HIPAA compliance in LLM-generated clinical summaries - achieving zero policy violations across 10K test documents. Directly applicable to RADV audit documentation and regulated healthcare reporting workflows.
PyTorchHuggingFaceLangChainSageMakerChromaDBRL / PPO
↗
[ 05 ] Finance
🤖
Autonomous Trading Agent - Reinforcement Learning
DQN-based RL agent achieving +15% return over baselines in rigorous backtesting. Deployed as a FastAPI REST API on AWS with Docker containerization - demonstrating reinforcement learning applied to sequential decision-making and real-time agentic AI architectures.
PyTorchTensorFlowOpenAI GymFastAPIDockerAWS
↗
[ 06 ] Healthcare
📈
Patient Outcome Prediction & Cohort Segmentation
LSTM/GRU discharge readiness models combined with K-Means cohort stratification across 6 patient segments - improving hospital resource planning accuracy by 18%. Full MLflow experiment tracking and Tableau reporting deliver end-to-end observability from model training to clinical insight.
PyTorchTensorFlowLSTM / GRUSageMakerMLflowTableau
↗
[ 07 ] Healthcare
🔬
Medical Imaging Classification - Computer Vision
ResNet/MobileNetV3 models classifying X-ray and MRI abnormalities at 92% accuracy, deployed via MLflow and Docker for real-time clinical inference. PySpark preprocessing at scale ensures the pipeline handles production-grade imaging volumes without bottlenecks - bridging research-quality CV with clinical deployment demands.
PyTorchTensorFlowOpenCVResNetMobileNetV3PySparkMLflowDocker
↗
