Shima Rashidi

I am a senior data scientist with a PhD in Computer Science, with expertise in MLOps, statistical modeling, deep learning, and pattern recognition within large-scale and commercial contexts. With a strong background in Python programming, I have worked extensively with cloud platforms to develop and deploy scalable data-driven solutions. I bring value to companies as a trusted data scientist who can take ownership of data driven projects from the initial pitch to the final delivery. I thrive in collaborative environments, working seamlessly across diverse teams, and take pleasure in innovating new features to enhance machine learning capabilities.

• Rashidi, Shima, and Saeed Sharifian. "A hybrid heuristic queue based algorithm for task assignment in mobile cloud." Future Generation Computer Systems 68 (2017): 331-345.
• Rashidi, Shima, et al. "Optimal visual search based on a model of target detectability in natural images." Advances in Neural Information Processing Systems 33 (2020): 9288-9299.
• Rashidi, Shima, et al. "An active foveated gaze prediction algorithm based on a Bayesian ideal observer." Pattern Recognition 143 (2023): 109694.
• Rashidi, Shima, et al. "IT-RUDA: Information Theory Assisted Robust Unsupervised Domain Adaptation." arXiv preprint arXiv:2210.12947 (2022).

Citizenship: Australian citizen

Working rights in UK: In the process of Global Talent Visa for United Kingdom

• Volleyball, Hiking, Biking, Camping, Redaing, Music

webcamsralia - 2024: Developed LLM traditional models for email/spam classification for intercepting emails at service desk level, topic generation for customer-agent webchats and their further classification via transfer learning (83% accuracy), and embedding models such as Ada, SBert and Bert.

AGL Australia - 2024: Refactoring legacy models to leverage PySpark and Databricks with a feature store enhances scalability, accelerates data processing, and fosters seamless collaboration, thereby unlocking the full potential of big data analytics for improved business insights. 

AGL Australia - 2023: Developed patterns for end-to-end MLOps, from data preparation to post-deployment through design and development of CI/CD toolkits as python packages as i) custom ML model auto-tunning using Mlflow wrapper, ii) model monitoring and iii) explainability using python design patterns (factory method, iterators). 

AGL Australia - 2023: Developed model quality assurance components through developing baselines as well as champion- challenger models for ML deployments in production    - Ensuring that all developed products follow PEP 8 Style guide

- Implementing testing pipelines for all products using Pytests 

AGL Australia - 2023: Developed and deployed churn models in production using LSTM and LGBM as a champion- challenger model with a robust accuracy of 86% and f1 score of 74. AGL Australia - 2023: Designed and deployed a multi-output regression model that harnessed 1d CNNs to capture local information from hourly data and LSTM to grasp long-term patterns This model accurately predicted EV charging patterns using customer smart meter data and was successfully adopted by one of AGL's retail companies. 

AGL Australia - 2023: Developed and deployed a time series forecasting model through designing a deep neural network with a mult`ihead attention layer (to capture the independent variances through a year) and a lstm head The model successfully forecasted the call center input calls with a normalized mse of 0.1 . 

RMIT University - 2022: Designed and developed a robust domain adaptation method (with Resnet50 as the backbone of the network) using a general measure from information theory employing deep learning, statistical analysis, and image processing, and assigning tasks to the data collection and development teams while utilizing Python (PyTorch, OpenCV, Torch-vision, Scikit-learn), Linux, Cuda toolkit, and Latex technologies.

Keylead Health - 2021: Detecting heart failure through voice analytics: Design and implement a deep neural network to classify patient’s voice for heart failure with an accuracy of 96% using Python (PyTorch, Scikit-learn), deep learning models (Resnet50, vgg16). Keylead Health - 2021: Medical emergency prediction: Design and development of a medical emergency prediction system based on tabular clinical data and electronic health records using deep learning achieving an accuracy of 93% using data wrangling techniques, feature selection, feature correlation, machine learning, Sql.

• Deep knowledge of Machine Learning (supervised/unsupervised, regressors, SVM, decision trees, clustering, transfer learning, lstm)
• Experienced with Natural Language Processing (LLM, Generative AI, Langchain, RAG, Transformers)
• Extensive experience in Deep Neural Networks (CNNs, attention, transformers), Statistics, probability theory, Optimization.
• Professional programmer in Python, R, C++, and MATLAB with coding experience in Linux.
• Extensive experience with common data science libraries (NLTK, Scikit-learn, SpaCy, OpenCV, SciPy, NumPy, Pandas) and machine learning frameworks (PyTorch).
• Experience in mining large and complex datasets using SQL (MySQL) and pySpark, dask.
• Extensive experience in development and deployment of machine learning models in Azure (Azure resources, blob storage, ML studio, VS code, monitoring, online and batch endpoints) and devops practices CI/CD.
• Experience with DataBricks, MLOps Stacks, Mlflow, feature stores (Azure and databricks)
• Experience with Vector databases (MangoDB, Faiss), open-ai services, logic apps, functions apps.

Soft skills: 

• Highly adaptable, agile person who can work autonomously, as part of a team, or as the leader of a team.
• Creative and analytical thinker, approaching problems with a novel perspective.
• Excellent verbal and written communication skills and able to foster solid relationships.