Software engineer · AI systems

I build AI systems for real work.

I’m interested in the layer between a capable model and a useful outcome: tools, state, permissions, evaluation, and the feedback loops that make the system dependable.

See projects Get in touch

Selected work

A few projects that represent what I build.

View all projects

Software, research, and AI systems.

Job postings to tailored résumés.

Can vision-language models solve physical puzzles?

For my master’s thesis, I built a benchmark that tests whether vision-language models can solve physical puzzles. I developed the simulation pipeline and five evaluation tasks, including an interactive challenge where models observe results and revise their actions. In the interactive evaluation, GPT-4o and Claude 3.5 Sonnet achieved roughly 5–7% success, with no systematic improvement across attempts.

Started Oct 2024

Language

Framework

Library

Database

Infrastructure

Toolkit

Tools and technologies I use across my work.

Language

Framework

Library

Runtime

Database

Cloud

Infrastructure

Integration

CLI

Protocol

Tool