Selected work

Systems that work. People who grow.

The résumé records roles and results. This page is the shorter story: six bodies of work that show how I approach difficult technical problems, build learning environments, and create programs that last.

Systems · Current

Probook

Reliable AI voice agents

As Chief Architect, I work on the full system around real-time AI: speech, model orchestration, tools, workflow state, observability, evaluation, and recovery. The central idea is simple: reliability is an architectural property, not a model feature.

Role: Chief Architect
Domain: Real-time AI for home services
Focus: From plausible response to completed task

Current focus →Architecture note →Probook ↗

People · 2016–Now

Columbia University

A competitive-programming community

Courses are only one part of a learning system. At Columbia, the work spans curriculum, practice, coaching, feedback, and community—creating an environment where students can steadily turn unfamiliar problems into tractable ones.

Scale: 429+ community members
Teaching: 14 course offerings through Fall 2026
Teams: Two ICPC World Finals appearances coached

Courses and evaluations →Team record →Mentorship record →

Institutions · Global

ICPC Foundation

A distributed technical-talent pipeline

The internship program connects students, mentors, universities, and technical projects across borders. The work is not merely placement; it is the operating system around selection, contribution, mentorship, continuity, and public accountability.

Completed: 136 successful internships
Current: 14 active interns as of August 14, 2026
Record: Public, searchable program archive

Program overview →Explore the directory →

Learning at scale

Google

Google Kick Start

During six years at Google, I worked in applied AI research and large-scale machine-learning infrastructure, and served as tech lead for Kick Start as it grew from a regional contest into a global competition. It was an early lesson in turning expert practice into an accessible, repeatable experience.

Role: Tech lead
Reach: 110K+ participants worldwide
Thread: Technical education at global scale

Career record →Problem-solving profile →

Craft · In progress

With Josh Alman

Competitive Programming: Beyond the Basics

A practical book about the space between knowing standard algorithms and solving unfamiliar problems fluently: modeling, pattern recognition, experimentation, proof, implementation, and the review habits that make practice compound.

Audience: Problem solvers beyond the introductory level
Subject: The craft behind consistent progress
Status: In development

Book project →Essay on deliberate practice →

Research · Foundations

Stanford · MIT · Google

Research across vision, networks, and programming systems

The domains changed, but the questions stayed recognizable: how to make complex computational systems more capable, observable, and useful. The publication record spans computer vision, networks, and programming languages.

Impact: 637+ research citations
Venues: CVPR, IJCV, OOPSLA, and others
Training: Stanford and MIT

Selected publications →Google Scholar ↗