Negin Ashrafi
PhD Student, Biomedical Data Science, Stanford University · she/her
About
I am a PhD student in Biomedical Data Science at Stanford University. My background is in machine learning, deep learning, reinforcement learning, and algorithms: dynamic programming, mathematical optimization (linear, integer, and nonlinear programming), and stochastic modeling. My methods work covers sequence models, self-supervised pretraining, knowledge distillation, active learning, and preference optimization (DPO), applied to clinical prediction and decision-making from large-scale EHR data. Before Stanford, I built production LLM pipelines as an AI/ML Scientist at TWG Global and developed deep learning models for ICU outcome prediction on MIMIC-III/IV at USC.
My current research is on building and evaluating AI systems for clinical decision-making, with an emphasis on safety. I design LLM-based agents for high-stakes clinical workflows and build the benchmarks and evaluation frameworks needed to test them. I also develop methods that detect and reduce clinically consequential failures: hallucinations, omissions, over-refusal, and recommendations that contradict the evidence. I currently work on the ARPA-H ADVOCATE program, developing rigorous, scalable evaluation methodology for patient-facing AI agents.
I am always open to interesting research collaborations and internship opportunities. If you have one in mind, feel free to reach out at ashrafi@stanford.edu.
- LLM agents
- AI safety & evaluation
- Benchmarks
- Clinical decision support
- Predictive modeling
- Reinforcement learning
- Optimization & algorithms
- Deep learning on EHR
Publications
- From Mutation to Medication: Characterizing Failures of Clinical LLMs to Respect Genomic ResistanceN. Ashrafi, R. DaneshjouMachine Learning for Health (ML4H) 2026
- OpenMTB-Audit: Exposing Over-Refusal and Clinical Expert Perspectives in LLM-Based Molecular Tumor Board Safety EvaluationN. Ashrafi, J. Luo, S. M. Frumm, R. DaneshjouPacific Symposium on Biocomputing (PSB) 2027
- HDLFORGE: A Two-Stage Multi-Agent Framework for Efficient Verilog Code Generation with Adaptive Model EscalationA. Abdollahi, N. Ashrafi, S. Shokoufa, M. Kamal, M. PedramNeurIPS 2026 Workshop on AI for Chip Design (AI4ChipDesign)
- TimingLLM: A Two-Stage Retrieval-Augmented Framework for Pre-Synthesis Timing Prediction from VerilogA. Abdollahi, N. Ashrafi, M. Kamal, M. PedramIEEE International Conference on LLM-Aided Design (ICLAD) 2026
- A Novel Multi-Task Teacher-Student Architecture with Self-Supervised Pretraining for 48-Hour Vasoactive-Inotropic Trend Analysis in Sepsis Mortality PredictionH. Jin, N. Ashrafi, K. Alaei, E. Pishgar, G. Placencia, M. PishgarIEEE Journal of Biomedical and Health Informatics
- Aligning Language Models with Clinical Expertise: Direct Preference Optimization for Heart Failure Nursing Documentation in Critical CareJ. Fan, L. Sun, N. Ashrafi, K. Alaei, M. PishgarComputer Methods and Programs in Biomedicine Update
- Auditing Construct Overlap in Explainable Machine Learning: Evidence from Burnout-Depression Prediction Across Student CohortsA. Dehghan, N. AshrafiarXiv:2607.10633
- Advanced Predictive Modeling for Enhanced Mortality Prediction in ICU Stroke Patients Using Clinical DataA. Abdollahi, N. Ashrafi, X. Ma, J. Zhang, D. Wu, T. Wu, Z. Ye, M. PishgarPLoS ONE
- Enhanced Prediction of Ventilator-Associated Pneumonia in Patients with Traumatic Brain Injury Using Advanced Machine Learning TechniquesN. Ashrafi, A. Abdollahi, K. Alaei, M. PishgarScientific Reports
- AI-Driven Solutions to Improve Safety and Health: Application of the REDECA Framework for Agricultural Tractor DriversN. Ashrafi, S. Yousefi, G. R. Aby, S. F. Issa, H. Darabi, K. Alaei, G. Placencia, et al.PLOS Global Public Health
- Optimized Feature Selection and Advanced Machine Learning for Stroke Risk Prediction in Revascularized Coronary Artery Disease PatientsY. Si, A. Abdollahi, N. Ashrafi, G. Placencia, E. Pishgar, K. Alaei, M. PishgarBMC Medical Informatics and Decision Making
- Optimizing Urban Mobility Through Complex Network Analysis and Big Data from Smart CardsL. Sun, N. Ashrafi, M. PishgarIoT
- Machine Learning-Based Prediction of ICU Readmissions in Intracerebral Hemorrhage Patients: Insights from the MIMIC DatabasesS. Chen, J. Fan, A. Abdollahi, N. Ashrafi, K. Alaei, G. Placencia, M. PishgarIntelligent Computing: Proceedings of the 2025 Computing Conference
- Data-Driven Machine Learning Approaches for Predicting In-Hospital Sepsis MortalityA. Shumilov, Y. Zhu, N. Ashrafi, G. Lian, S. Ren, M. PishgarIntelligent Computing: Proceedings of the 2025 Computing Conference
- LLM Encoder vs. Decoder: Robust Detection of Chinese AI-Generated Text with LoRAH. Jin, N. Ashrafi, A. Abdollahi, W. Liu, J. Wang, G. Gui, M. Pishgar, H. FengarXiv:2509.00731
- A Machine Learning-Based Prediction of Hospital Mortality in Mechanically Ventilated ICU PatientsH. Li, N. Ashrafi, C. Kang, G. Zhao, Y. Chen, M. PishgarPLoS ONE
- Prediction of Sepsis Mortality in ICU Patients Using Machine Learning MethodsJ. Gao, Y. Lu, N. Ashrafi, I. R. Domingo, K. Alaei, M. PishgarBMC Medical Informatics and Decision Making
- Prediction of 30-Day Mortality for ICU Patients with Sepsis-3Z. Yu, N. Ashrafi, H. Li, K. Alaei, M. PishgarBMC Medical Informatics and Decision Making
- Deep Learning Model Utilization for Mortality Prediction in Mechanically Ventilated ICU PatientsN. Ashrafi, Y. Liu, X. Xu, Y. Wang, Z. Zhao, M. PishgarInformatics in Medicine Unlocked
- Process Mining / Deep Learning Model to Predict Mortality in Coronary Artery Disease PatientsN. Ashrafi, A. Abdollahi, G. Placencia, M. PishgarInternational Conference on Computers and Industrial Engineering (CIE51)
- Prediction of In-Hospital Mortality for ICU Patients with Heart FailureJ. Zhang, H. Li, N. Ashrafi, Z. Yu, G. Placencia, M. PishgarInternational Conference on Computers and Industrial Engineering (CIE51)
- Optimizing Mortality Prediction for ICU Heart Failure Patients: Leveraging XGBoost and Advanced Machine Learning with the MIMIC-III DatabaseN. Ashrafi, A. Abdollahi, J. Zhang, M. PishgararXiv:2409.01685
- Effect of a Process Mining based Pre-processing Step in Prediction of the Critical Health OutcomesN. Ashrafi, A. Abdollahi, G. Placencia, M. PishgararXiv:2407.02821
Full list on Google Scholar.
Education
PhD, Biomedical Data Science2025 – PresentStanford University · Stanford, CACoursework Data Science in Medicine · Representations and Algorithms for Computational Molecular Biology · Overview of Biomedical Data Science · Design and Analysis of Algorithms
MS, Analytics2022 – 2024University of Southern California · Los Angeles, CACoursework Machine Learning · Foundations of AI · Text Analytics · Predictive Analytics · Optimization for Data Science · Stochastic Processes · Design of Experiments · Quantum Algorithms and Optimization
BS, Industrial Engineering2018 – 2022Sharif University of Technology · Tehran, IranCoursework Data Mining · Probability Theory · Engineering Statistics · Linear Algebra · Computer Simulation · Operations Research I & II · Programming Languages
Experience
Graduate Research Assistant2025 – PresentStanford University · Department of Biomedical Data ScienceEvaluation of patient-facing AI agents (ARPA-H ADVOCATE) · LLM-as-a-judge and clinician rubrics · agentic auditing frameworks · hallucination and omission detection · active learning for pathology
AI/ML Scientist2025TWG Global HoldingsDPO-based prompt optimization · production LLM document pipelines
Graduate Research Assistant2022 – 2025University of Southern California · Industrial and Systems EngineeringSelf-supervised pretraining and knowledge distillation for ICU time series · DPO alignment for clinical documentation · multi-agent LLM frameworks · clinical risk prediction on MIMIC-III/IV
- Machine Learning Engineer Intern2024Stealth Startup
End-to-end ML pipelines for sales and demand forecasting
Undergraduate Research Assistant2021 – 2022Sharif University of TechnologyData mining and A/B testing for customer modeling