Vision

The kind of systems I want to keep building.

I am most interested in AI work that helps people move through difficult information more clearly, and in the engineering choices that make that experience reliable.

Useful beats flashy

I care more about whether a system helps someone think or act better than whether it produces an impressive one-minute demo.

Evaluation is part of design

Retrieval, ranking, and interaction quality should be measured early, not treated as cleanup work after the model is chosen.

Knowledge should feel navigable

I am drawn to interfaces that help people move through dense information with confidence instead of friction.

Research should survive contact with reality

I like turning research ideas into systems that can handle messy documents, ambiguous questions, and real constraints.

What I am aiming toward

Bridging research ideas and everyday usefulness.

The work I want to do sits in the space between model capability and human clarity. That means retrieval systems that can be trusted, evaluation loops that expose real weaknesses, and interfaces that help people understand rather than just generate. I do not want AI systems to feel magical from a distance and confusing up close.

Over time, I want to get better at building tools that make knowledge more navigable: systems for reasoning over documents, graph-structured information, and messy real-world questions that do not arrive in clean benchmark form.