SolEvolve: LLM-driven Evolutionary Discovery of Algorithms
An LLM-guided evolutionary search system for discovering and testing coding-theoretic constructions.
An autonomous search system where LLM-guided operators, SAT-seeded search, and verification loops rediscover strong coding-theoretic constructions.
problem
Algorithmic discovery in coding theory needs a loop that can propose candidates, test them, preserve evidence, and continue searching without relying on one-shot generation.
key idea
Combine LLM-guided mutation/recombination, SAT-seeded search, and verification loops to explore construction spaces.
my role
Lead researcher and system builder.
methods
- • Evolutionary search
- • SAT-seeded candidate generation
- • LLM-guided operators
- • Verification loops
evidence / results
- • Submitted/manuscript track
- • Acts as a precursor to later benchmark-evolution and formal-agent systems
why this belongs in the portfolio
- • Turns search trajectories into reusable research evidence
- • Connects coding-theory search with agentic experimentation
authors
Jae-Hyun Baek
venue / status
Master's thesis / manuscript track
Submitted manuscript / thesis-derived research line.
tags