I build the thing — not just advise on it. A decade charting the terrain between boardroom strategy and production code: terabyte pipelines for one of the world’s great banks, ML systems for Fortune 500s and federal agencies at a global consultancy, and evaluation research at the frontier of AI.
I operate where most maps go blank — the contested ground between corporate strategy and hands-on engineering. I've directed the analytics on a flagship program, shipped ML that governs a great technology house’s revenue book, and stress-tested frontier LLMs for published research.
The route wasn't drawn in advance. Midway through a computer science degree at Holy Cross, the fund behind one of my scholarships faltered and the total award fell 76%. In two weeks I engineered a credit bridge across two institutions, kept researching, and graduated on time — a year later I was writing open-source scientific software at Harvard. That instinct, finding structure in chaos, became the through-line: from a great bank's data infrastructure to a global consultancy's hardest client terrain, to the frontier-AI work I do today.
Off the grid: Greek-American dual citizen, eight seasons of competitive rowing (Head of the Charles included), a violinist, and a habit of painting faint stars in dark landscapes — a reminder to navigate by weak signals when clarity is scarce.
Bachelor's Computer Science
Won Mellon Foundation funding to found my own research team building a public-transit accessibility app. When the fund behind one of my scholarships faltered mid-degree — cutting the total award 76% overnight — I engineered a two-institution credit bridge and still graduated with my class.
ASPIRE Clinical Researcher
Diagnostic-accuracy research on septic arthritis in an emergency-medicine lab, shadowing the ER at Barnes-Jewish Hospital. The summer that started an obsession: watching software, models, and data change real-world outcomes.
Software Engineer, Data Provenance
Co-authored open-source scientific software (provViz, DDG-Explorer) that captures and visualizes the full lineage of research data — making scientific results reproducible and verifiable.
Big Data Engineer
Built terabyte-scale ELT infrastructure consumed across the asset-management business — Spark and Hadoop pipelines, NLP applications, automated reporting. Led a firm-level digital modernization and managed an intern team shipping a data-lake monitoring dashboard.
Analyst → Acting Expert Engagement Manager
Six years on the expert track, leading cross-functional teams of engineers, data scientists, and PhDs to ship production-grade data products for Fortune 500s and federal agencies — culminating as Project lead of analytics on a flagship program, with a disaster-resilience study delivered to Harvard under McKinsey's Global Managing Partner.
Founding Technologist, stealth venture · LLM evaluation researcher
Founding technologist for a stealth AI venture — LLM-enabled financial planning, core simulation architecture. Domain expert for Mercor Intelligence evaluating frontier models across the leading labs, contributing to the published APEX-Agents benchmark.
Several volumes of this record remain sealed at the request of their commissioning parties. Where a seal appears, the account is complete but closed.
As project lead, I built a constraint-based production simulation modeling unit-level output against inventory and capacity limits to expose the binding bottlenecks hidden across thousands of components.
Clustering over quote, transaction, and market data for one of the great technology houses — segmenting deals to surface over-discounted cohorts, then converting insight into pricing guidelines and a sales-governance playbook.
Integrated quote, transaction, and market data into a master dataset; ML clustering segmented deals by industry, region, and sales behavior, surfacing over-discounted cohorts. Insight became mechanism: KYC-driven pricing guidelines and a governance playbook that aligned sales incentives with margin optimization.
Combinatorial simulation of next-generation product platforms: a components catalogue spanning 10T+ theoretical configurations, filtered by constraints to a locally-executable set, scored on four mission KPIs, and mapped to a Pareto frontier.
A two-model architecture with constraint-based filtering collapsed one component space to millions of configurations and a second from billions — ensuring compatibility and local executability. Each surviving configuration was scored on four mission KPIs, plotted on an interactive cost-performance scatter, and Pareto-frontier analysis prioritized the platforms worth committing capital to.
Rebuilt a public agency's XGBoost fraud classifier from the ground up as fraud behavior shifted week to week — engineering hundreds of features across behavioral signals, velocity metrics, and network anomalies to flag high-risk cases in near real-time.
Features spanned behavioral signals (suspicious fund routing), velocity metrics (non-human completion times), network anomalies (geolocation and time-of-day mismatches indicating VPN use), and profile inconsistencies — flagging high-risk cases in near real-time as fraud behavior, data, and policy shifted week to week.
Conceived and led an OCR + NLP + regex pipeline that turned 100K+ unstructured contracts into trusted, queryable data assets — deployed across multiple enterprise clients.
Tiered entity resolution across a global supplier's purchasing data — deterministic SKU matching first, NLP text-similarity fallback — unlocking cross-enterprise comparison and a ~5% savings rate across an exchange-scale procurement book.
Tiered entity resolution: deterministic SKU matching for high-fidelity linkage, NLP text-similarity as the fallback for items without unique IDs, and QC prioritized by spend volume on lower-fidelity matches — unlocking accurate cross-enterprise comparison and actionable sourcing recommendations.
Domain expert designing professional-services benchmarks that expose failure modes in frontier models — structured RLHF feedback contributing to the published APEX-Agents evaluation research.
Designed evaluation tasks, rubrics, and grading criteria for complex professional-services scenarios — treating evaluation design and prompt structure as a coupled system — and provided structured RLHF feedback on model behavior and failure modes.
Years of deployed client work in each territory — from sealed flagship programs in the Marches to guild-house pricing engines to published AI research. The craft itself — software, data, the instruments — runs through all fourteen years.
I take on engagements where strategy and engineering have to arrive together — data products, AI systems, and the teams that ship them.
Connect on LinkedIn→Most consultants I know agree the new machines are about to reshape the trade — yet little of how the trade operates has meaningfully changed. I left the great firms to be one of the ones who changes it.
The research-and-analysis layer that once took a team of analysts two weeks is becoming something these machines catalyze in hours. The pyramid was not built for this moment. I'd argue it is about to become an obelisk.
— from the chronicler's notes, Vol. XIV
