Evaluation-first RAG system with dual retrieval baselines (BM25 + FAISS), production FastAPI service, comprehensive testing (239 tests), drift monitoring, and full ML lifecycle automation.

Domain-adapted LLM for structured travel guidance with evaluation-first design, synthetic instruction tuning, and offline inference.

Fast, maintainable portfolio built with Next.js App Router, MDX for content management, Tailwind CSS styling, and file-based architecture for easy updates.

Explainability and uncertainty estimation for medical image classification with CNNs. Evaluation-first approach treating confidence and explanations as measurable outputs.