Data Scientist & ML Engineer · Toronto, ON
Negin Amirsoleimani

Negin Amirsoleimani

I build ML pipelines and LLM systems that move business numbers.

4+years in applied ML
60%faster data processing
8–10%revenue uplift
500K+records modeled
View projectsGet in touch
system architecturerunning
  1. inputPDF upload
  2. processDocument processing
  3. retrieveInformation retrieval
  4. modelZephyr-7B inference
  5. outputGrounded answer
Retrieval-Augmented Generation for Document Intelligence →
// about

From data to decisions

Data scientist with 4+ years building machine learning models and analytical solutions that improve business operations. Experience designing ETL pipelines, developing predictive models, and analyzing data to support strategic decisions. Skilled in Python, SQL, and cloud platforms (AWS, Azure) with strong communication abilities to translate technical work for non-technical audiences. Committed to delivering data-driven insights that drive measurable results.

// experience

Where I've shipped models

January 2026 - Present

Data Scientist

York University · Toronto, CA

  • Built end-to-end Python pipelines to clean, combine, and prepare data from sources, including drone images, satellite data, and sensor records.
  • Developed and compared ML models (Random Forest, XGBoost) with statistical models (mixed-effects, GAMs, Bayesian spatiotemporal models) to find patterns across space and time and measure the impact of an intervention.
  • Checked and explained model results using SHAP, uncertainty analysis, leave-one-group-out cross-validation, and ablation studies.
  • Processed geospatial and satellite imagery with ArcGIS Pro, Google Earth Engine, and WebODM to create useful indicators from raw image data.
  • Created visualizations and a dashboard to present results clearly to researchers and stakeholders and support data-driven decisions.
February 2026 - August 2026

Machine Learning Researcher – 3D Shape Modelling and Mesh Reconstruction

Western University · Toronto, CA

  • Developed an end-to-end 3D machine learning pipeline for pediatric ear shape prediction, combining statistical shape modeling, point-cloud deep learning, and 3D mesh reconstruction.
  • Improved point-level prediction accuracy by 27% through optimized training and combined adult-pediatric shape modeling.
  • Built data preprocessing, model training, evaluation, and 3D reconstruction workflows in Python for point-cloud and mesh-based data.
  • Evaluated 3D geometries using Chamfer distance, statistical validation, and mesh-quality analysis, visualizing surfaces in MeshLab.
February 2025 - December 2025

Applied Data Scientist

Cyber Pattern Inc · Toronto

  • Designed and deployed end-to-end ML pipelines leveraging advanced causal inference techniques (econometric modeling, panel data analysis, regression) on 500K+ records to evaluate operational patterns and deliver data-driven strategic recommendations.
  • Engineered scalable ETL pipelines integrating transactional databases, APIs, and external data sources, reducing data preparation time by 35% and enabling 25% operational efficiency gains and 8-10% revenue uplift through cross-functional partnership with senior leaders.
December 2020 - November 2022

Machine Learning Engineer

Tanesh Parsa · Mazandaran, Babol

  • Built and deployed predictive models and optimization algorithms using machine learning techniques across 50+ SQL databases to analyze user behavior patterns, increasing click-through rates and driving measurable engagement improvements and 8-10% advertiser revenue growth.
  • Designed automated ETL processes using Python, SQL, and Power Query, replacing manual Excel workflows; ETL processes reduced data processing time by 60% and minimized human error.
// selected work

Technical projects

01
Zephyr-7Bdocument-grounded answers

Retrieval-Augmented Generation for Document Intelligence

  • Developed an end-to-end RAG system using Python, Streamlit, and Zephyr-7B for context-aware question answering over PDF documents.
  • Integrated document processing, information retrieval, and LLM inference to generate responses grounded in retrieved document content.
  • Built an interactive Streamlit interface supporting PDF uploads, natural-language queries, and document-grounded responses.
Generative AIRAGLLMsPythonStreamlit
View code
02
Llama 3adapted to a specific domain

Domain-Adaptive Large Language Model Fine-Tuning

  • Fine-tuned Meta Llama 3 using Hugging Face tools for domain-specific text generation and language understanding tasks.
  • Developed data preparation, model training, and evaluation workflows to adapt pretrained language models to specialized applications.
  • Evaluated model adaptation and inference to investigate response relevance and performance on domain-specific tasks.
Generative AIHugging FaceLlama 3Python
View code
03
R² 0.803on 3,460 resume–job pairs

Resume–Job Compatibility Prediction Using GNNs and LLMs

  • Developed a graph-based recommendation system integrating GAT, GraphConv, and OpenAI embeddings for semantic resume-job matching.
  • Constructed heterogeneous knowledge graphs from resumes, job descriptions, and skills using Python, spaCy, and PyTorch Geometric.
  • Trained and evaluated graph-based models, achieving R² = 0.803 on 3,460 resume-job pairs and outperforming TF-IDF and conventional LLM-based approaches.
Graph Neural NetworksLLMsPythonPyTorch Geometric
View code
// stack

Technical skills

Machine Learning & Causal Inference

Supervised/Unsupervised LearningPredictive ModelingRegression AnalysisEconometric ModelingPanel Data AnalysisFeature Engineeringscikit-learnXGBoostLightGBM

Programming & Data

PythonpandasNumPySQLWindow FunctionsCTEsQuery OptimizationRGit/GitHub

Data Engineering & ETL

ETL/ELT PipelinesData Quality AssuranceData ModelingWorkflow AutomationDatabricks

Cloud & MLOps

AWS SageMakerAzure Data FactoryAzure Data LakeML Model VersioningCI/CD PipelinesDocker

Statistical Analysis

Hypothesis TestingA/B TestingTime-Series ForecastingStatistical InferenceExperimental Design

AI Tools & Automation

LLMs (Claude, ChatGPT)Prompt EngineeringAI-Assisted DevelopmentAutomated Feature Engineering

Data Visualization

Power BITableauMatplotlibSeabornKPI TrackingStakeholder Communication
// research

Publications

IEEE ISSCS 2025

Optimizing ML Pipelines for Classification & Predictive Analytics

Engineered automated ML pipeline analyzing 50GB+ of sensor data, achieving 99% classification accuracy through advanced preprocessing, feature extraction, and model optimization techniques.

Published Research

Optimization Modeling for Resource Allocation & Operational Efficiency

Developed optimization model for resource allocation in distributed systems, achieving 15% improvement in scheduling efficiency and significant cost reduction through multi-criteria decision-making.

// education

Education

January 2023 - February 2025GPA 3.9/4

Master of Applied Science

Electrical & Computer Engineering

University Of Windsor · Windsor, Canada

Statistical AnalysisData MiningMachine LearningAdvanced AnalyticsDatabase SystemsSignal ProcessingOptimization
Completed

Bachelor of Science

Computer Engineering (Software)

Mazandaran University of Science and Technology · Mazandaran, Babol

Database DesignData StructuresAlgorithmsStatisticsProbability TheorySoftware Engineering
// contact

Have a project or role in mind?

The fastest way to reach me is email. I'm also on LinkedIn and GitHub.

neginamirsoleimani@outlook.comLinkedInGitHub