SolEvolve
An agentic search loop for proposing, mutating, testing, and preserving mathematical candidates.
An LLM-guided evolutionary discovery system for algorithmic and mathematical search, developed around coding-theory constructions.
problem
One-shot generation is a poor fit for discovery; useful mathematical search needs memory, operators, constraints, and repeated falsification.
why it matters
The same search-and-verify pattern appears in coding theory, benchmark generation, and formal proof workflows.
maker note
This is the part of my work where the agent is allowed to be a messy researcher: propose, mutate, test, fail, and try again.
what I built
- • LLM-guided evolutionary search pipeline
- • SAT-seeded candidate generation direction
- • Verification-oriented research traces
evidence
- • Master’s thesis/manuscript track
- • Connects to coding-theory construction search and EntropyMath-style lineage
current status
Submitted/manuscript track; used as a bridge between coding-theory search and agentic evaluation.
next step
Clarify what part of the trajectory can become reusable data for specialized mathematical agents.
role
Lead researcher and system builder.
tags
artifact links
Code