Sareh Soltani Nejad
Data Scientist / Machine Learning Researcher
BrainsCAN
About Me
I’m a Data Scientist and Machine Learning Specialist with over 5 years of hands-on experience developing scalable ML solutions and extracting insights from complex, high-dimensional datasets. I am passionate about creating impactful solutions, especially in the healthcare and finance sectors, and I am always keen to learn and grow in areas that merge ML and real-world applications.
My professional interests are broad yet interconnected, including machine learning, computer vision, Biomedical Computing, AI-driven healthcare solutions, and everything related to LLMs and natural language processing (NLP). I thrive at the intersection of technology and real-world applications, continually seeking innovative solutions that make a meaningful impact. I am currently looking for my next job opportunity in machine learning.
Personal email: sarehsoltani.inbox@gmail.com
Work email: ssolta7@uwo.ca
Interests
- Machine Learning
- Applied ML in Healthcare
- Computer Vision
- Biomedical Computing
- Natural Language Processing (NLP)
Education
MSc in Computer Science
Western University, Canada
BSc in Computer Engineering
Amirkabir University
Experience
Data Scientist - AI Engineer
BrainsCAN
- Collaborated on the large-scale OMMABA project to explore the impact of music perception on brain functionality and develop a multimodal dataset integrating behavioral, EEG, and fMRI data from 60 participants.
- Developed a robust data preprocessing pipeline to enhance data quality, consistency, and usability for downstream analysis.
- Achieved 94% accuracy in ECG arrhythmia classification by developing deep learning models (1D CNNs, RNNs) and traditional ML algorithms (SVM, XGBoost), with statistical insights into model performance.
- Deployed a production-ready pipeline via Dockerized FastAPI, enabling scalable, real-time arrhythmia detection.
- Annotated immune cell types from 20K single-cell RNA-seq blood samples using generative variational autoencoders (VAE) and scGPT (transformer-based model), leading to improved accuracy and interpretability.
- Tools: Pandas, Scikit-learn, TensorFlow, HuggingFace, scanPy, sciPy, Seaborn, Optuna, MLflow, FastAPI, Docker, AWS
Machine Learning Engineer Technical Facilitator
Vector Institute for AI
- Conducted two cohorts of Anomaly Detection Bootcamp as part of the Vector ML Experts team.
- Deployed tailored ML solutions to address anomaly detection use cases for 12 companies by collaborating with stakeholders.
- Developed an ML-based fraud detection framework using ensemble methods (LightGBM), boosting model accuracy by 27% and reducing false positives by 15%.
- Implemented a DL-based TabNet model for financial fraud detection, achieving 88% accuracy on transaction data.
- Built a scalable video anomaly detection framework using multiple instance ranking, reaching an AUC of 85%
- Tools: Docker, GCP, Pandas, SQL, PySpark, Scikit-learn, PyTorch, Streamlit, T-Test, Matplotlib, Tableau, Wandb, Git, Slurm
Data Scientist Intern
Vector Institute for AI
- Led an Anomaly Detection Workshop for 50+ professionals, delivering hands-on training on advanced techniques.
- Delivered a reference fraud detection demo using a credit card dataset, achieving 93% AUC with an AutoEncoder.
- Conducted analysis for a pharma company by integrating lab data from 1,000 patients to assess their drug impact on BMI and blood pressure, applying statistical, subgroup, and outlier analyses with external data to optimize clinical trial design.
- Leveraged supervised models (Random Forest, XGBoost) to estimate treatment effects, predicting a 12% reduction in BMI and a 3% reduction in blood pressure.
Machine Learning Research Engineer
Western University
- Designed a novel weakly-supervised video-anomaly detection system built on a two-stream I3D ConvNet.
- Built a data pipeline to process 1TB+ of video data from UCF-Crime benchmark, leveraging Multiple Instance Learning.
- Automated extraction of appearance (RGB) and motion (optical - flow) embeddings through parallel two- stream I3D encoders
- Devised a late-fusion strategy that improved accuracy by 20%, achieving an 85% AUC and surpassing published baselines.
- Tools: Python, PyTorch, OpenCV, Weights & Biases, Matplotlib, Git
Machine Learning Engineer
IPM & Sharif Brain Center
- Applied NER and topic modeling to extract structured insights from unstructured EHR doctors’ notes, streamlining patient records and reducing manual chart review time by 30%.
- Used large language models like BERT for clinical treatment categorization.
- Developed 3D medical imaging visualizations from CBCT data, improving diagnostic accuracy for 50 patients.
- Tools: Python, Pandas, LLM, HuggingFace, BERT, PyTorch, NLTK, spaCy
Projects
Built a real-time ECG heartbeat classification system using CNNs trained on the MIT-BIH dataset. The model automatically extracts key temporal ECG features and was optimized with Optuna, with experiments tracked via MLFlow for robust performance. Deployed as a scalable REST API on AWS EC2, the system accepts CSV inputs and delivers fast, accurate predictions, achieving 95% accuracy in classifying arrhythmias. This end-to-end solution enables cardiologists and researchers to analyze their own data in real-world healthcare settings, improving clinical diagnosis through efficient, scalable arrhythmia detection. Code
Immune Cell type Annotation from scRNA-seq Blood Data Using scGPT | In collaboration with UHN
Built an end-to-end transformer-based pipeline to classify ~20K blood cells from scRNA-seq data, identifying ~10 immune subtypes using scGPT embeddings, clustering, and marker gene validation. Preprocessing included HVG selection and canonical immune marker validation. Results highlight the potential of large-scale generative models for high-resolution immune profiling. Code
This project introduces a cutting-edge approach to anomaly detection in urban surveillance systems using Two-Stream Inflated 3D (I3D) Convolutional Networks. By capturing both spatial and temporal features more effectively than traditional methods, our model significantly improves detection precision. Leveraging a weakly supervised Multiple Instance Learning (MIL) framework, we treat surveillance videos as collections of ranked clips, enabling efficient anomaly identification with minimal manual labeling. Optimized for real-world deployment, this scalable and high-performing solution sets new standards in public safety technology through intelligent, context-aware video analysis. Code.
