Dexterous Data

Data science consulting for real-world results

The Problems I Solve

Your model works on test data but fails in production. Benchmark accuracy doesn’t guarantee real-world performance. I help teams design validation strategies that test what actually matters for their use case — not just whether the numbers look good on held-out data.

Your team is building ML features without formal training. Smart engineers can ship models. Knowing whether those models are doing what you think they’re doing requires a different skillset — experimental design, statistical reasoning, and validation methodology. I build custom training materials and workflows tailored to your team’s stack and problems.

You hit a wall you can’t debug past. Whether you’re a researcher whose analysis pipeline broke, a developer whose ML integration isn’t behaving, or a founder prototyping a data feature — sometimes you need someone who’s seen the problem before. I do hourly sessions for focused problem-solving.

How I Work

Most engagements start with a conversation about your specific situation. From there, work typically falls into one of three shapes:

Project-based consulting. A defined scope — build a validation pipeline, audit a model’s methodology, design a data collection strategy. Clear deliverables, clear timeline.

Team training. Custom materials for your stack and your problems. Not generic “intro to ML” workshops — targeted content that addresses the gaps your team actually has. I’ve built an entire open-source learning platform as proof of concept for this approach.

Hourly problem-solving. You’re stuck, you need a second set of eyes, and you need it this week. Code review, debugging, architecture decisions, methodology questions.

Why Me

I have a PhD in psychology with 15+ years of research experience in neuroimaging and applied data science. That background means I think about data problems from the methodology up — not just “does the code run” but “does this approach actually answer the question you’re asking.”

My open-source work demonstrates how I think and build:

Face Value — An end-to-end computer vision pipeline showing that domain-matched weak supervision outperforms benchmark datasets on real-world validation. Complete with an interactive results dashboard.

Pixel Process — An open-source data science platform with applied case studies, workflow patterns, and interactive learning materials. The full source code is public.

Get In Touch

If any of this sounds relevant to what you’re working on, send me an email and tell me what you’re dealing with. I’ll let you know if I can help.