Reliability under pressure.

Measured & reproducibleAI evaluation

Measuring whether a persona survives real use

A reproducible evaluation program for identity, voice, conversational continuity, and drift across changing models and longer interactions. The public story is about measurement design, not the underlying product implementation.

Scope

Multi-model evaluation, attribution, drift, and conversational-flow checks

Discipline

Fixed cohorts, honest baselines, confidence intervals, and reproducible reports

Result

Found the operating regime where pooled evidence is useful, and documented where it is not

What stays private: product-specific prompts, live infrastructure, customer workflows, private datasets, and deployment configuration.

Production decision-makingOperations research

Choosing the simplest defensible system

Operational systems rarely optimize for one variable. The work is to turn an ambiguous production question into an explicit decision across quality, cost, risk, implementation time, and future flexibility without allowing sophistication to become the goal.

Frame

Translate product urgency into measurable technical and operational constraints

Compare

Evaluate the quality-cost frontier under representative and adverse conditions

Decide

Ship the least complex approach that clears the operating bar, then optimize from evidence

Public boundary: the method is described here; the product, operating domain, measurements, infrastructure, and implementation remain private.

Lifecycle disciplineTechnical leadership

Ending systems responsibly

Technical ownership does not end when active development stops. A responsible closeout makes costs visible, retires unused infrastructure, assigns what remains, preserves useful knowledge, and leaves the organization with an explicit operating state.

Question

What should stop, what should remain, and who owns each surviving obligation?

Discipline

Close infrastructure, documentation, access, cost, and communication loops together

Outcome

A deliberate transition rather than an indefinitely lingering project

Public boundary: this describes the lifecycle practice without identifying the project, systems, people, costs, or internal decisions.

Submitted + in preparationWith Josh Alman

Cascade centrality and network augmentation

A research program asking how influence changes when a network gains an edge, combining exact hardness, efficient approximation, structural laws, augmentation counterexamples, and exact rational computation.

Submitted

Exact hardness and efficient approximation · ITCS 2027

In preparation

Structural decomposition and nonmonotone network augmentation

Evidence

Proofs, explicit counterexamples, exact enumeration, and independently checkable computation

Release disciplineICPC problem development

Engineering a contest that fails safely before launch

High-stakes programming contests need more than clever problems. Their statements, validators, generators, accepted solutions, rejected solutions, and data must agree under adversarial scrutiny.

Scale

17-problem regional draft plus qualifier and regional validation work

Checks

Input rejection, independent solutions, wrong-answer and time-limit baselines

Principle

Automation raises the floor; independent human sign-off remains the release gate

Integrity boundary: no unreleased statements, solutions, test data, or contest-sensitive material appears here.

Operating at scaleICPC Foundation

Designing programs that can outlast their builders

Across ICPC Foundation Internships and ICPC University, I am building the systems around technical learning: how people enter, contribute, receive feedback, find the next useful resource, produce public work, and eventually take ownership themselves.

Internships

Real technical work across websites, archives, infrastructure, testing, mobile, and training

ICPC University

Goal-based routes across courses, practice, lectures, camps, events, and contest operations

Direction

Formal charter, shared ownership, explicit KPIs, documented workflows, and durable handoffs

Institutional test: the work is successful when quality, access, and accountability survive changes in leaders, mentors, and sponsors.