Siavash Soltani
Vancouver,BC
Summary
Lead Data Scientist with a proven track record at Beatdapp, driving significant improvements in fraud detection rates through innovative ML models and generative AI applications. Expert in Graph Neural Networks and LLMs, skilled in cross-functional collaboration, delivering impactful insights and enhancing operational efficiency.
Overview
2
2
years of professional experience
Work History
Lead Data Scientist
Beatdapp
03.2024 - Current
- Led a team of five data scientists and analysts in developing and deploying production-grade ML models for music streaming fraud detection.
- Drove a 40%–200% increase in fraud detection rates for key clients, including one of the top five largest music streaming services, by ideating novel features and optimizing detection algorithms.
- Pioneered the company's first use of generative AI by developing and deploying a Gemini-based LLM to automate artist information gathering, replacing manual Google searches, and saving significant analyst time and the company’s resources.
- Designed and led the development of a novel fraud claim detection model, analyzing artist metadata for another top-five industry client to identify fraudulent content claims.
- Innovated by employing Graph Foundation Models to detect sophisticated streaming fraud across massive datasets, generating high-quality embeddings for downstream classification tasks.
- Collaborated cross-functionally with Product, Data Engineering, and Full Stack teams, and led weekly client meetings to deliver actionable insights via client-facing dashboards.
Postdoctoral Fellow
University of British Columbia
06.2023 - 02.2024
- Conducted independent research projects to advance the understanding of complex atomistic systems.
- Achieved a 10x increase in rare event identification efficiency in atomistic simulations by applying ML-based statistical analysis, published in a high-impact journal.
- Constructed and deployed Graph Neural Networks (GNNs) to predict time-series behavior, and used CNNs for dynamical mode decomposition.
- Primary Tools: Python, TensorFlow, Scikit-Learn, Keras, SQL, and Pandas.
Education
Ph.D. - Materials Science (Machine Learning Focus)
University of British Columbia
Vancouver, CA07-2023
Skills
- Machine learning models
- Fraud detection algorithms
- Generative AI applications
- Data analysis techniques
- Graph Neural Networks (GNNs)
- Large Language Models (LLMs)
- Graph Foundation Models
- Time-Series Analysis
- Statistical Modeling Tools & Stack: Python (Pandas, NumPy, Scikit-Learn, Keras, TensorFlow), SQL, Git
- Leadership: Team Leadership (5 reports), Client-Facing Communication, Project Management, Cross-Functional Collaboration (Product, Engineering)
Timeline
Lead Data Scientist
Beatdapp
03.2024 - Current
Postdoctoral Fellow
University of British Columbia
06.2023 - 02.2024
Ph.D. - Materials Science (Machine Learning Focus)
University of British Columbia