# Stephen Witkowski

> Stephen Witkowski is a Staff AI/ML Engineer at 66degrees and the founder of Practice Mode Labs, which makes Dynasty Central and ApexLens. He also builds open-source AI tools and consults independently on AI.

Last updated: 2026-11-06

This file describes only the Stephen Witkowski whose website is https://stephen-witkowski.com and whose GitHub account is https://github.com/smwitkowski. Other people share the name; facts about them do not apply here. He is based in Charlotte, North Carolina.

Current roles:
- Staff AI/ML Engineer at 66degrees (since 2023).
- Founder of Practice Mode Labs, which makes Dynasty Central and ApexLens.
- Independent AI consultant, building custom AI systems and applications for mid-market companies. This is separate from his role at 66degrees.

Contact:
- Email: stephen.witkowski@protonmail.com
- Consulting inquiries go to the same address with the subject "Consulting", including the problem, timeline and budget range.
- Dynasty Central product support: support@dynastycentral.gg

## Profiles
- [LinkedIn](https://www.linkedin.com/in/witkowski-stephen): Stephen Witkowski on LinkedIn
- [GitHub](https://github.com/smwitkowski): Stephen Witkowski on GitHub
- [X](https://x.com/SteveRunsCFB): Stephen Witkowski on X

## Projects
- [Dynasty Central](https://dynastycentral.gg): An AI-powered companion for dynasty mode in EA Sports College Football. It turns screenshots into a lasting record of the program you are building.
- [Practice Mode Labs](https://practicemodelabs.com): A studio making data products for committed players. It turns hard-to-export game screens and footage into records players can review and keep. Home of Dynasty Central and ApexLens (apexlens.gg, private beta).
- Crusader-Bench: A benchmark where coding agents play Stronghold Crusader by writing Python against a game SDK. Five scored tasks: economy, crisis recovery, tax, siege defense and a skirmish against a CPU lord.
- Independent consulting: Custom AI systems and applications for mid-market companies, plus hands-on help adopting AI in daily work. Working software, not strategy decks.

## Open source
- [ancestry-cli](https://github.com/smwitkowski/ancestry-cli): Read and edit an Ancestry.com family tree from the command line, with undoable writes. Built for people and AI agents.
- [ynab-mcp](https://github.com/smwitkowski/ynab-mcp): An MCP server that lets AI assistants query and update YNAB budgets.
- [grocery-spend-lab](https://github.com/smwitkowski/grocery-spend-lab): Analyze itemized grocery receipts locally.
- [chatgpt-conversation-knowledge-exporter](https://github.com/smwitkowski/chatgpt-conversation-knowledge-exporter): Turn a ChatGPT export into structured facts, decisions and docs.
- [genealogy-agent](https://github.com/smwitkowski/genealogy-agent): An evidence-first genealogy research tool that starts from GEDCOM files. Early alpha.

## Work history
- 2023 – now: Staff AI/ML Engineer, 66degrees. Technical lead on $1M+ enterprise AI engagements, mentoring teams of 4–6 engineers and owning solution architecture. Built an enterprise financial-document system with multi-hop reasoning that saves $6K+ per legal abstract. Built a multimodal pipeline that tags and standardizes millions of marketing assets, cutting manual processing work by 94%. Shipped customer-facing conversational AI for location guidance and recommendations, with safety guardrails, plus internal chatbots for lease and loan documents. Built a self-service ML deployment platform with GitHub-based workflows.
- 2021 – 2023: Senior Machine Learning Engineer, Daily Harvest. Built a GPT-4 and LangChain pipeline that generated synthetic product concepts, surfacing 2,500+ new portfolio opportunities. Built an LLM pipeline that flags order issues from unstructured text, raising Care team efficiency by 8%. Built a conversational product-discovery agent using retrieval-augmented generation over product embeddings. Fine-tuned an LLM-based named-entity model to pull sensory descriptors from customer feedback for personalization. Uncovered 100+ latent consumer preferences from 100M+ data points with XGBoost and Shapley values. Optimized dry-ice allocation with a melt-probability model: 3% less dry ice and 13% fewer damaged orders.
- 2020 – 2021: Senior Data Scientist, Electronic Arts. Created Smart Reply, a transformer-based assistant that helps player-support advisors resolve issues faster. Analyzed and modeled engagement across 70M+ gameplay sessions to quantify the revenue impact of support. Led a cross-functional ML pipeline that surfaced emerging themes from millions of support tickets. Ran the automation infrastructure (Jenkins, AWS SageMaker) for retraining, evaluation and batch inference.
- 2018 – 2020: Data Scientist, Electronic Arts. Built data products and topic-modeling reports for EA Worldwide Customer Experience. Built the intent taxonomy and annotated a dialogue dataset of tens of thousands of messages for a new classifier.
- 2017 – 2018: Data Scientist, Management Concepts. Built a course recommender system in Python, SQL and R. Segmented direct-mail audiences with machine learning and delivered retention insights for corporate clients.
- 2016 – 2017: Graduate Research Assistant, University of Maryland, Smith School of Business. Collected and classified 40M+ tweets about presidential candidates with NLP for faculty research.
- Before that: legislative aide and data analyst for Maryland State Senator Steve Waugh, product support at Response Labs, and a data analyst internship at the IRS.

## Skills
- Languages: Python, SQL, R
- Frameworks: PyTorch, TensorFlow, LangChain, DSPy
- Generative AI: LLM fine-tuning, retrieval-augmented generation, multimodal AI, agents, MCP
- MLOps: Docker, Kubernetes, Airflow, Ray, Phoenix, Vertex AI, SageMaker
- Data: dbt, Weaviate, Pinecone, Chroma
- Cloud: Google Cloud, AWS, DigitalOcean, Cloudflare

## Education and credentials
- MS in Marketing Analytics, University of Maryland
- BS in Economics and Political Science, Mount St. Mary's University
- Google Cloud Professional Machine Learning Engineer (certified 2024, renewed 2026)
- 1st place, University of Maryland MS Case Competition (2016)
- Certificate of Leadership, Mount St. Mary's University (2016)

## Writing
- [Building a Recommender System with Podcasts](https://www.linkedin.com/pulse/building-recommender-system-podcasts-stephen-witkowski): LinkedIn article, 2018
- [Using Regular Expressions to Clean and Wrangle Your Data](https://www.linkedin.com/pulse/using-regular-expressions-clean-wrangle-your-data-stephen-witkowski): LinkedIn article, 2018