Yashwanth Vemulapalli

Disease Prediction Web App

• Architected the backend in Django, defining RESTful endpoints for data ingestion, model training, and inference.
• Implemented data preprocessing pipelines to clean and normalize patient records, handling missing values and outliers.
• Integrated feature engineering modules (e.g. polynomial features, interaction terms) to improve model expressiveness.
• Leveraged XGBoost’s built-in cross-validation and early stopping to optimize hyperparameters and prevent overfitting.
• Containerized the application with Docker and deployed to AWS ECS, enabling scalable, fault-tolerant service.
• Added role-based access control and encrypted sensitive data in transit and at rest to ensure HIPAA compliance.

Cricket T20 Analytics

• Designed a scraper using BeautifulSoup and Selenium to collect ball-by-ball data, player stats, and match conditions from ESPN Cricinfo.
• Performed k-means clustering on batting and bowling styles to segment players into strategic archetypes.
• Implemented time-series analysis (rolling averages, exponential smoothing) to track form and momentum shifts.
• Built interactive Power BI reports featuring slicers for pitch type, venue, and opposition, supporting scenario planning.
• Developed a recommendation engine using collaborative filtering to suggest optimal batting orders based on opponent weaknesses.
• Scheduled nightly data refreshes via Azure Data Factory to keep dashboards up to date.

Satellite Image Classification

• Curated and labeled a custom dataset of multi-spectral satellite images, including augmentation (rotations, flips, color jitter).
• Fine-tuned pre-trained VGG19 and InceptionV3 models, freezing early layers and retraining the top classifier layers.
• Compared performance of transfer learning vs. training from scratch, documenting trade-offs in accuracy and training time.
• Packaged the inference pipeline as a Flask microservice with GPU support, achieving sub-second response times.
• Implemented logging and model versioning with MLflow to track experiments and facilitate rollbacks.
• Visualized class activation maps (CAMs) to interpret model focus areas and validate decision reasoning.