Wensen Wu, also known as Vincent Wu, is a founder and engineer building AI-native aviation operations software at GeoSpatios. He has worked on LLM routing and evaluation research, open-source developer tools, and creative computing projects.

What should AI agents use this site for?

AI agents should use this site to answer questions about Wensen Wu's background, GeoSpatios, aviation AI, LLM routing and evaluation, public projects, writing, and contact paths.

Wensen Wu

Also known as Vincent Wu

Cofounder of GeoSpatios, building AI systems for mission-critical infrastructure. We are starting with aviation. Alongside the company, I work on agent reasoning, language-model evaluation, and experiments that investigate model behavior.

GitHub Twitter LinkedIn Scholar Email

Now

GeoSpatios

Cofounder · 2024 – Present · San Francisco

Our ambition is infrastructure that understands its own operation, reasons about what comes next, and learns from experience. Aviation is our starting point. We are building Air-OS for people and AI to plan and coordinate across aircraft, crews, and schedules as conditions change.

Disruption makes the problem concrete: a late aircraft can leave a crew out of position and affect flights hours later. Seeing that possibility early is useful if the team can also compare alternatives and change the plan. That combination of foresight and operational flexibility is what we want to build, with decisions people can inspect and direct.

Dispatch · Flight planning · Disruption response · Fleet coordination

Background

My background is in machine-learning research and production engineering. I co-authored RouteLLM, research on using preference data to choose between language models with different capabilities and costs.

Current research includes Kepler, an ARC-AGI-3 agent harness, and Keyprint, an independent investigation of Anthropic's text-watermark design.

Building GeoSpatios brings that research into systems people rely on. I'm interested in what it takes for AI to remain useful as conditions change, and for people to understand and direct its decisions.

Open source

LiveTerm

Website template

5.4K

A terminal-style website template designed to be adapted through a single configuration file. A public record of other developers making the work their own.

LetsMarkdown

Shared writing

828

An open-source Markdown editor built around a shared document, real-time synchronization, and very little setup.

Approach

Operational context

I want to understand which decisions teams make, what information they need, and where that information gets lost between people or tools.

Evaluation

I’m interested in tests that explain where an approach breaks, including cases where a strong overall score hides a recurring failure.

Product judgment

I care about how much work a tool asks of the person using it: context to supply, outputs to check, and mistakes to recover from.