Designed a multi-agent orchestration system using CrewAI and FastAPI, implementing a directed workflow that coordinates specialized GPT-4o-mini agents for real-time travel preference extraction and itinerary generation
Engineered deterministic data handoffs using Pydantic validation and custom Python tools for budget allocation, producing consistent financial calculations through structured context filtering and validation.
Developed a PyTorch-based CNN for binary melanoma classification using 10K dermoscopic skin-lesion images, applying image preprocessing, normalization, and data augmentation techniques including random flips and rotations to improve model generalization.
Trained and evaluated the model on separate training and test datasets, achieving strong test accuracy and validating performance using precision, recall, F1-score, confusion matrix.
