Cassandra Sullivan

About

ML/NLP practitioner working across healthcare and business/finance use cases: currently fine-tuning a domain-specific Llama 3.1 model for safe reasoning over EHRs/clinical notes using RAG + RL/DPO and RAGAS-based evaluation. Has built UMLS-driven entity normalization pipelines with quantified quality gains and developed embedding/vector-DB systems (FAISS) for semantic matching and forecasting/recommendation applications at Aurora AI and Banxico.

Key Strengths

• Built healthcare LLM inference/analytics pipeline combining RAG + RL with clinician-feedback reward model to reduce hallucinations
• Designed multi-stage biomedical data cleaning + UMLS-based entity normalization; reduced duplicate entities by 18% and increased UMLS alignment by 12%
• Validated entity resolution at data/semantic/model levels (manual gold set + downstream fine-tuning impact)
• Improved downstream factual accuracy/consistency by ~9–10% using cleaned vs raw datasets
• Production-grade Python workflow design with orchestration, versioning, and experiment tracking (Airflow/Prefect, DVC, MLflow)
• Embedding + vector DB search/relevance systems (OpenAI/Sentence-Transformers + FAISS) linking customer behavior and product metadata
• Clear prototype-to-production decisioning using measurable metrics (accuracy, latency, interpretability) and cross-functional validation

Experience

Education

Awards

• Data Science Institute Merit Scholarship - University of Chicago
• Academic Excellence Award (Top 5%) - Instituto Tecnológico y de Estudios Superiores de Monterrey
• Magna Cum Laude - University of Chicago
• Magna Cum Laude (94/100) - Instituto Tecnológico y de Estudios Superiores de Monterrey

Skills

Languages