Joining Scale Venture Partners as Chief Data Scientist

I’m excited to share that I’ve joined Scale Venture Partners as Chief Data Scientist. After spending the last few years building AI/ML systems to extract insights from venture data at Aumni and JPMorgan Chase, I’m now working directly with a VC firm to embed those capabilities and more into their investment process.

The Opportunity

Scale operates in what we call the “messy middle” of venture: between pure seed investing and late-stage traction plays. It’s the stage where pattern recognition matters, but the patterns are subtle. Where timing is everything, but the signals are noisy. Where the best founders often break conventional molds, making traditional screening approaches insufficient.

The firm has been thoughtfully building toward this for a while. They’ve published extensively about their vision for AI in venture capital, and what struck me most was their clarity on what not to automate. As they put it in their recent post: we’re looking for “empathy for how investors think, rigor on how models are designed, and taste in knowing what not to automate.”

That resonates deeply with my experience building ML systems in complex domains that humans actually need to use.

How I Got Here

Looking back, my career has been a surprisingly consistent thread of building AI systems for domains with limited data and high stakes.

It started with predictive maintenance at Predikto, where we tried to predict when industrial assets would fail using sparse sensor data and limited failure history. The challenges were similar to what I’ll face at Scale: long time horizons, rare events, and users who needed to understand why the model flagged something, not just that it did.

At Raytheon Technologies, I led applied data science across defense and aerospace systems, domains where explainability isn’t optional and getting it wrong has consequences. You learn quickly that the best model isn’t always the one with the highest AUC; it’s the one your users trust enough to act on.

At Aumni, I built the data science function focused on extracting insights from venture capital legal documents. We processed millions of pages to understand deal structures, valuations, and terms. It gave me a deep appreciation for how messy venture data really is and how much signal there is if you know where to look.

The common thread: building ML systems that augment expert judgment rather than replace it. Systems that surface the right information at the right time, while remaining interrogable and grounded in explainable features.

It’s a messy middle of it’s own, a mixture of machine learning, applied AI, research, application development, and consulting.

What We’re Building

At Scale, we’re building infrastructure to help our investors discover opportunities earlier and make better informed decisions. This breaks down into a few key areas:

People as Signal: Tracking when exceptional talent joins early-stage teams. In venture, who decides to join is often more revealing than what gets pitched. We’re building systems to identify these career endorsements systematically.

Hiring Intelligence: Going beyond simple hiring velocity to understand who companies recruit and where they come from. There’s a difference between generic growth and strategic talent acquisition that signals product-market fit.

Predictive Round Timing: The goal is to anticipate when companies will raise, not just react after rounds close. This is one of the persistent frustrations in venture.

Market Pattern Detection: Identifying clusters of similar startups emerging within short timeframes as early indicators of category formation and market shifts.

Lots and lots more: We’re capitalists in the business of making better decisions, so the landscape of data, models, and applications that we will build to do so is wide.

The technical challenge is interesting: unlike domains with abundant training data (fraud detection, ad targeting, recommendation systems), venture has limited historical examples and long feedback loops. Success depends on identifying outliers and founders who break molds: exactly the cases where pattern matching struggles.

The cultural challenge is equally important: the best infrastructure has to be both accurate and usable. In venture, “usable” means investors can interrogate it, pressure test it, and gut check it. It means prioritizing explainability over marginal gains in predictive accuracy. It means building judgment support systems, not decision automation.

That’s the philosophy Scale has articulated, and it’s what drew me to this role.

What’s Next

Over the coming months, I’ll be embedding into the investment process, understanding how decisions get made, and identifying where data and models can genuinely help.

If you’re working on similar problems (AI for expert decision-making, ML with sparse data, explainable models for high-stakes domains), I’d love to connect.

Here’s to the next chapter.