Suggestions
Mark Conway
Director of AI | Software Builder | Data Scientist | Machine Learning Engineer
Mark Conway is a highly skilled data scientist and machine learning expert currently working in a stealth mode position.1 His LinkedIn profile provides insights into his extensive experience and expertise in the field of artificial intelligence and data science:
Professional Experience
Current Role: Mark is currently working as a Data Science Consultant at Ahold Delhaize USA in Chicago, Illinois.1 In this role, he has been focusing on:
- Implementing Large Language Models (LLMs) for semantic search and substitution
- Developing Learning-To-Rank (LTR) models for personalized customer recommendations
- Creating an NLP Evaluation Toolkit for search and substitution
Previous Positions:
- Head of Machine Learning at Scottfree Analytics LLC (May 2017 - Present)
- Principal Data Scientist at Deloitte (September 2021 - July 2022)
- Data Science Consultant at FIS (January 2020 - July 2021)
- Senior Data Scientist at Walmart (July 2014 - May 2017)
Key Achievements
-
Open-source Contribution: Mark is the sole developer of AlphaPy, an AutoML framework for trading and sports, which has been downloaded over 132,000 times.12
-
Consulting Experience: He has provided data science consulting services to Fortune 500 and international clients through Scottfree Analytics LLC.1
-
AI-driven Retail Platform: At Deloitte, Mark defined business requirements for an AI-driven retail shopping platform, incorporating multiple models for various analytical purposes.1
-
Synthetic Data Innovation: While at FIS, he persuaded the executive team to sell Synthetic Data, leading to $75M in potential contract deals.1
Technical Expertise
Mark's skillset includes:
- Machine learning and deep learning techniques
- Natural Language Processing (NLP)
- Large Language Models (LLMs)
- AutoML frameworks
- Big data technologies (e.g., Spark, Hadoop)
- Causal inference and A/B testing
Mark Conway is described as a "rare breed of data scientist" who combines superior statistical and mathematical acumen with technical expertise.1 His diverse experience across multiple industries and his contributions to open-source projects demonstrate his commitment to advancing the field of artificial intelligence and data science.
Highlights
Over ten years ago, our CEO Robert Scott recorded NBA odds and results by hand in his engineering notebook. The notebooks piled up over the years, and it was time to automate. As a Data Scientist, you can learn a lot about models through empirical observation first. And sports models are hard, especially with modern-day lines derived through machine learning. Money Line models are easily calibrated because of winning percentage, but then you have to hurdle Expected Value, based on the difference between predicted and implied probability. Spread models are even harder because the calibration is usually lost, combined with a relative paucity of training data (NFL: 272 games per year). Finally, Over/Under models are a coin flip.
So, Robert decided to put seven machine learning models to the test, ranging from XGBoost to Logistic Regression. On a daily basis, these models track the NBA, NFL, MLB, NHL, and NCAA football and basketball. The classification targets are:
- Won On Points
- Won on Spread
- Over
The models have been running since October 2023, and the overall results are not surprising in some cases. But here are some teasers:
- Is betting the Away and Under a generally good strategy?
- Do model outcomes run in streaks, and is there reversion to the mean?
- Can you fade consistently bad models because of overfitting?
If you would like to explore on your own, please follow the link in the first comment.
When my colleague Robert Scott and I started doing Kaggle competitions about 10 years ago, we automated the feature engineering, model selection, and visualization components that eventually became the AutoML library AlphaPy. To date, the package has > 134K downloads and over a thousand stars. We open-sourced it on a whim, and the next version AlphaPy Pro is a fully automated sports and markets RAGtime framework. Stay tuned after all the eggnog has been GulPT!
