Sushma Mundlamuri

Designed and deployed ML models (XGBoost, LightGBM) to predict aircraft component supply chain disruptions. Processed multi-source datasets including supplier, logistics, and real-time inventory data using Snowflake and Python. Engineered features from news articles and ERP data with TF-IDF, NER, and embeddings. Developed ETL workflows via Apache Airflow to maintain data freshness and model retraining. Incorporated SHAP explainability into risk predictions for stakeholder transparency. Delivered visual insights on Tableau to procurement teams. Automated anomaly detection and alerts. Collaborated with supply chain strategists and analysts to align models with business SLAs and mitigation strategies. Applied time-series forecasting and reinforcement learning principles for inventory management and allocation optimization.

Collected and preprocessed multimodal MRI and PET imaging data ensuring consistency for deep learning pipelines. Engineered CNN and autoencoder-based pipelines for Alzheimer’s disease progression biomarkers. Applied transfer learning with ResNet and EfficientNet for improved early detection. Fine-tuned GPT-4 to generate structured radiology summaries. Built RAG pipelines with FAISS and ChromaDB for clinical question answering. Deployed models with FastAPI, Flask, and Streamlit dashboards. Managed experiment tracking and containerized deployments on Google Cloud Vertex AI with CI/CD. Collaborated with clinicians for validation and compliance, ensuring HIPAA adherence. Visualized model features using attention heatmaps and provided weekly technical syncs to cross-functional teams.

Built ML pipelines for real-time risk scoring of transactions integrated within decisioning platforms. Developed scalable REST APIs with Flask and deployed containerized models on Kubernetes clusters for auto-scaling. Integrated streaming telemetry and logs through Apache Kafka. Collaborated with data engineers on feature store development for fraud detection models. Maintained live model monitoring using Prometheus and Grafana with drift detection alerts. Managed MLflow-based experiment tracking and robust audit trails. Partnered with backend, fraud, and compliance teams to ensure smooth deployments adhering to finance industry standards. Led A/B testing, holdout validation, and forensic analysis for false positives and model improvement.