Machine Learning

Samira Pakravan

Work Authorization: US Permanent Resident

Cupertino, CA, 95014 575-***-**** # ***********@*****.*** Google Scholar ï linkedin § github

Education

University of California, Santa Barbara Sep 2018 – Jun 2024 Ph.D. in Mechanical Engineering, emphasis in Computational Science & Engineering (CSE) Thesis: Hybrid Machine Learning Algorithms for Solving Forward and Inverse Problems in Physical Sciences New Mexico State University Aug 2013 – Dec 2015

M.S. in Computer Science

Sadjad Institute of Higher Education Sep 2008 – Sep 2012 B.S. in Computer Science

Experience

Genentech, Roche May 2024 – Aug 2024

Intern - Generative AI with LLM Clinical Pharmacology - Artificial Intelligence & gCS Foundational AI

• Designed & deployed an AI-driven chatbot to empower scientists in transforming raw clinical data into analysis-ready datasets, streamlining workflows, enabling faster, more accurate data analysis tailored to clinical pharmacology needs.

• Built the co-pilot using Retrieval-Augmented Generation (RAG), function-calling agents, LlamaIndex, LangChain, Azure OpenAI (GPT & Ada models), Streamlit, PandasAI, and rpy2 to automate complex data transformations.

• Enhanced accessibility for non-programmers by enabling conversational interactions for dataset creation, modification, and analysis through tailored prompts, reducing reliance on coding expertise.

• Collaborated with domain experts to align chatbot functionality with clinical pharmacology requirements and iteratively improved its usability through user feedback and testing.

• Delivered significant productivity gains, reducing the time from raw data to actionable insights, and presented the impact to stakeholders, demonstrating enhanced efficiency and flexibility for scientific workflows. Genentech, Roche Sep 2023 – Mar 2024

Intern - AI for Drug Discovery & Development Clinical Pharmacology - Artificial Intelligence & gCS Foundational AI

• Developed a PyTorch-based machine learning software package to train predictive models using digital health data, enhancing clinical development processes.

• Designed a Pharmacology-Informed Neural-Stochastic Differential Equation (SDE) framework to model stochastic digital health data, capturing the underlying dynamical system.

• Improved insights into disease progression using generative AI algorithms, including Neural ODE/SDE, latent SDE, RNN

& GRU, applied to longitudinal clinical and simulated datasets.

• Characterized patient-to-patient variability through learned patient-dependent parameters, enabling personalized modeling of disease progression and treatment effects.

• Delivered a robust, reusable software tool for researchers to explore drug effects and patient variability, advancing the development of targeted therapies.

UC Santa Barbara Jan 2018 – Mar 2018

Student Intern Computational Applied Science Laboratory (CASL)

• Developed graphical user interface (GUI) for a scientific HPC software using QT. USDA-ARS Jornada Experimental Range Jan 2017 – Jan 2018 Mobile Application Developer New Mexico State University

• Developed a mobile application to detect the soil color using AngularJS, Apache Cordova, JavaScript.

• Designed and maintained the user interface using a HTML, CSS, JQuery. Tech Parks Arizona Jun 2016 – Oct 2016

Database Driven Web Application Developer University of Arizona

• Developed a website front-end and SQL database of compiled content using ASP.Net.

• Designed and maintained website using HTML, CSS, JavaScript, JQuery. Projects

JAX-DIPS, Differentiable Interfacial PDE Solver Python, JAX, Docker, Linux 2020 - Present

• Exploring hybrid approaches combining traditional numerical algorithms with deep learning for solving PDEs. BiPDE, Scientific Machine Learning Tensorflow, Keras, Linux 2019 - 2021

• We developed the “Blended Inverse-PDE Networks” that combine traditional methods for numerical computations of PDEs with modern deep learning architectures to discover hidden fields in data. BIPDE-Nets seamlessly incorporate domain-knowledge about physics of the problem.

Generating Molecules using NLP Models PyTorch, rdkit, Numpy 2023

• I created an NLP model to predict the next letter in a given SMILES string given the previously typed letters. The output is the most likely valid molecules similar to what it is queried. Computer Vision with Uncertainty Quantification PyTorch, OpenCV, Numpy, Docker 2023

• I applied image augmentation techniques on various datasets and trained the U-net model for image segmentation task. Monte Carlo dropout is used to estimate model uncertainty. Technical Skills

Deep Learning Models: Hands-on experience with GNN, CNN, RNN, GRU, LSTM, ResNet, Transformers GenAI Algorithms: Experienced with LLMs, Graph Neural Networks, VAEs, and Diffusion models Languages: Python, C++, MATLAB, Java, SQL

Platforms/Libraries: JAX, PyTorch, NumPy, Matplotlib, OpenCV, Scikit-learn, Pandas, Tensorflow, Keras, RDKit Technologies/Frameworks: Linux, Git, Docker, Kubernetes, Azure, AWS, Conda, Visual Studio Code, SVN, Rstudio Journal/Conference Publications

From Noise to Signal: Unveiling Treatment Effects from Digital Health Data through Pharmacology-Informed Neural-SDE ICLR 2024 Samira Pakravan, Nikolas Evangelou, Maxime Usdin, Logan Brooks, James Lu Learning From Time Series for Health JAX-DIPS: Neural bootstrapping of finite discretization methods JCP 2022 P Mistani, S Pakravan, R Ilango, F Gibou, Journal of Computational Physics Neuro-symbolic partial differential equation solver NeurIPS 2022 P Mistani,S Pakravan, R Ilango, S Choudhry, F Gibou Machine Learning and the Physical Sciences(ML4Phys) Solving inverse-PDE problems with physics-aware neural networks’ JCP 2021 S Pakravan, P Mistani, MA Calvo, F Gibou Journal of Computational Physics

( : equal contribution)

Teaching Experience

Teaching Associate (main instructor) Jun 2021 – Sep 2021 UC Santa Barbara Department of Computer Science

• I was main instructor for CMPSC16 - Problem Solving with Computers I (C++). Teaching Assistant Jan 2018 – Jun 2024

UC Santa Barbara Departments of Mechanical Engineering & Engineering Sciences

• ME 125/225ML - Special Topics: Machine Learning (Graduate Course, 2 times)

• ENGR 3 & ME 17 - Introduction to Programming & Mathematics of Engineering with Matlab (6 times)

• ME 152B & 151B - Fluid Mechanics II & Thermosciences II Teaching Assistant Aug 2014 – Dec 2015

New Mexico State University Department of Computer Science

• CS 271 - Object Oriented Programming with C++ (3 times). Teaching Associate/Research Assistant May 2014 – Jul 2016 Young Women in Computing (YWiC) New Mexico State University

• YWiC offers activities for K-12 students, showcasing computing creativity through hands-on and project-based learning. Conference Talks

ICIAM 2023 Tokyo, Japan

Minisymposium on deep learning, preconditioners, and linear solvers

• On learning neural operators of PDEs with interfacial jump conditions for accelerating simulations of physical systems Outreach / Extracurricular

• Grace Hopper Celebration Student Scholarship, USA, 2021

• Travel award for 6th ACM International Conference on BuildSys, Columbia University, New York, USA 2019

• Outstanding instructor award during Young Women in Computing summer camp, NMSU 2015

• Travel award for Graduate Cohort Workshop(CRA-W), San Francisco, USA 2015

• New Mexico State University International Alumni Scholarship, NM, USA twice 2013 & 2014 Outreach / Extracurricular

Peer Review 2019 – Present

• Journal of Computational Physics (9 manuscripts) Served as Judge 2016

• 27

th

annual NM Supercomputing Challenge for middle and high school students, NMSU, USA

Contact this candidate