Mara Hassan

Designed data architecture for a pharmaceutical regulatory intelligence platform supporting document search, information retrieval, and AI-powered workflows. Engineered data processing workflows using Python, SQL, and Google Cloud to organize structured and unstructured regulatory information. Built RAG workflows with semantic chunking, embeddings, and vector search, plus multi-agent workflows for document drafting, retrieval, citation verification, and response-quality checks. Evaluated retrieval precision, citation quality, hallucination risk, and response reliability. Improved data traceability and quality checks to help users verify information against source documents and collaborated with technical and business stakeholders to deliver practical data/AI solutions, reducing preparation time by ~45% and improving evaluation/iteration quality by ~30%.

Architected and implemented AI workflows for pharmaceutical regulatory documents, including ingestion and transformation, retrieval and semantic search, analytics, and automated knowledge work. Built NLP pipelines for large volumes of unstructured regulatory content using Python, BigQuery, Pub/Sub, Cloud Storage/Cloud Run, and Vertex AI. Designed semantic chunking, embeddings, and vector search to improve retrieval quality for downstream LLM applications. Developed evaluation frameworks for retrieval precision, hallucination rate, agent performance, citation accuracy, and response quality, enabling measurable production standards. Implemented citation verification and automated document-prep workflows to improve output quality and reduce document preparation time. Partnered with business and technical stakeholders to translate ambiguous regulatory needs into production AI architecture and scalable services.

Designed and deployed production AI workflows for pharmaceutical regulatory applications, including document processing, information retrieval, drafting, citation verification, and quality assurance. Built RAG pipelines using Vertex AI with semantic chunking and vector search to improve retrieval quality and support reliable AI-generated responses across large regulatory document collections. Developed LLM-powered document intelligence workflows to extract, organize, and retrieve information from complex pharmaceutical documents, and created evaluation pipelines to measure retrieval precision, hallucination rates, and agent performance for continuous testing and improvement. Improved document preparation time by 25% by automating ingestion and workflow stages, and improved AI output quality by 15% via retrieval optimization and evaluation-driven iteration. Built event-driven AI workflows using Python, SQL, Firestore, Pub/Sub, and Cloud Run for scalable production processing.