Afeefa Zaman

Designed a detection system using anomaly detection (Isolation Forest) and graph networks, flagging numerous suspicious transactions with 98% precision. Scaled credit risk models using TensorFlow and H2O.ai, processing 10TB of transactional data to reduce default prediction errors. Containerized ML workflows with Docker and Kubernetes, reducing model deployment time from 2 weeks to 3 days for cross-functional teams. Built a real-time recommendation engine (collaborative filtering) boosting user engagement by 25% for a fintech app. Automated data drift detection using Evidently AI, improving model retraining efficiency and maintaining 90%+ accuracy post-deployment. Collaborated with DevOps to migrate ML pipelines to Azure ML, cutting cloud costs while improving scalability.

Built a prediction model using XGBoost and PyTorch, analyzing 2M+ records to reduce hospital readmission rates. Automated feature engineering pipelines with Apache Spark, cutting preprocessing time for real-time inference workflows. Deployed a BERT-based NLP system to classify clinical notes, achieving 94% F1-score and reducing manual chart reviews by 70% for medical staff. Optimized AWS SageMaker inference costs using model pruning and ONNX runtime for high-volume prediction APIs. Led A/B testing framework development, enabling data scientists to validate 12+ models in production and iterate 3x faster. Mentored 4 junior engineers on MLOps best practices (CI/CD, model monitoring), slashing deployment failures.

Developed a claims prediction model (Random Forest + SHAP) to identify billing anomalies. Created NLP-powered chatbots (Rasa + SpaCy) to handle 80% of routine queries, reducing call center load. Streamlined ETL pipelines using Python and SQL, cleaning and standardizing records for analytics teams. Ran hyperparameter tuning experiments with Optuna, improving model accuracy by 12% across 8 core prediction tasks. Visualized model performance metrics in Tableau dashboards, adopted by executives for quarterly strategy reviews.