Selected Work

Problem: A preclinical hardware team had a working product and enthusiastic early users and was preparing to scale — without pressure-testing whether unit cost could reach a price the market would pay.

Approach: Went past the demo to the economics — true cost to build at volume vs. what real prospective buyers could budget, informed by the people who’d have to purchase it.

Deliverables: A feasibility and cost-reality memo with a clear go/no-go and the assumptions that would have to change for viability.

Outcome: The team entered its scale decision understanding the economic gap before committing further capital.

Problem: A team training models on a large library of behavioral video had inconsistent labels across annotators, quietly undermining every downstream model.

Approach: A multi-model QA pipeline — LLM-assisted auditing to flag disagreements/errors, plus specialized classifiers so experts reviewed only ambiguous cases.

Deliverables: Label-quality audit, QA pipeline, evaluation metrics, deployment documentation.

Outcome: Reduced manual review burden and improved label consistency, giving downstream models a foundation they could trust.

Problem: An investor was evaluating an early-stage company whose performance claims looked strong but hadn’t been independently scrutinized.

Approach: Reconstructed the scientific lineage, traced each claim to the actual evidence, assessed the regulatory path against what the data supported.

Deliverables: A decision-ready memo identifying which claims held, which were unsupported, and the questions to ask before committing.

Outcome: Surfaced evidence gaps that materially changed how the investor viewed the opportunity.

Problem: A founder preparing to raise had promising preclinical results built on a study design that wouldn’t withstand a sophisticated investor’s technical review.

Approach: Rebuilt around the decision it needed to support — endpoints tied to the claim, pre-specified analysis, proper powering, blinding/randomization, controls for the variables reviewers probe.

Deliverables: Revised study design and analysis plan + a clear account of what the existing data could and couldn’t claim.

Outcome: The team went into diligence with a package built for the questions it would actually face.

Problem: A research team wanted to “use AI” and was about to commit engineering time and a hire to an ambitious build without a clear read on where automation would pay off.

Approach: Mapped real workflows against where modern AI delivers vs. stalls; separated high-leverage, low-risk wins from expensive distractions.

Deliverables: A prioritized assessment of what to automate first, defer, or leave alone — with the reasoning.

Outcome: The team redirected effort to what would move the needle and avoided sinking months into a build that wouldn’t return the investment.

Problem: A team had invested heavily in a capable internal research system but hadn’t shaped it into something a customer could buy and use.

Approach: Worked across science and engineering to define what a market-ready version required — the gap between a working internal tool and a product, and the path to close it.

Deliverables: A product definition and technical roadmap grounded in what the technology could do and what the market would pay for.

Outcome: A concrete path from research asset to commercial product, with technical and economic realities accounted for up front.