Online evaluation, tracing, and audit trails to monitor inputs/outputs, model scores, country context, and prompt versions for safety reviews.
A nurse-facing AI assistant that benchmarks emergency classifiers, scores response quality, and queues doctor-reviewed labeling for continuous improvement.
Transformed large knowledge banks into structured, localized Q&A with a country-aware vector store and metadata tagging.
Created ~500 labeled emergency vs non-emergency messages and a doctor-reviewed benchmark for response quality.