01 / Parallel discovery
Many agents,
shared knowledge
I’m interested in how independent searches can share tools, infrastructure and useful discoveries — without simply repeating the same work.
Agents and research
How I approach research with agents.
I study how agents can ask useful questions, design experiments and revise their methods as they learn. I’m particularly interested in what happens when many searches can build on each other’s work.
What I work on
From more experiments
to better research.
Many agents,
02 / Self-improvement
Improve the method,
as well as the candidate
I study how agents diagnose failed experiments, revise search procedures and transfer useful discoveries to the next research task.
Selected publication
EvoQuantLab
EvoQuantLab: Training-Free Dual-Loop Evolution for Auditable Quantitative Research
As a co-author, I’m interested in the paper’s question of how a research procedure can itself become something an agent learns to improve. For me, that connects the search for useful candidates with the harder question of whether the method is becoming more reliable.
Read on arXivA research principle
I want improvements that survive independent tests.
I treat a promising result as a hypothesis to test again. I want to assess changes on evidence that was not used to develop them, and make the experiments clear enough for someone else to examine. That is how I want to distinguish a better research method from a fortunate outcome.
Work on these questions with me.
If you’re interested in self-improving agents, large-scale experimentation or research infrastructure, I’d like to hear from you.