funsearch

A scalable multi-node asynchronous implementation of DeepMind’s FunSearch using RabbitMQ.

A Python package for Large Language Model-Based Automatic Heuristic Design

Companion page for "Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery." A curated four-axis index covering MWPs, LLMs and reasoning models, multimodal geometry, Lean theorem proving, and verified discovery (FunSearch, AlphaEvolve, Erdős problems).

An LLM writes trading strategies as code; a search loop calibrates each one and stress-tests it out of sample against a luck baseline, surfacing the few that hold up. Runs on any Yahoo Finance ticker.

Finding better kubernetes schedulers using funsearch with an accelerated simulator

Verifier-gated evolutionary discovery: only proven candidates survive. Zero dependencies.

FunSearch-style evolutionary program discovery for the L-shape Ramsey problem

Autonomous neuro-symbolic research agent synthesizing, verifying, and sandboxing algorithms with FunSearch, CEGIS (Z3), Lean 4 proofs, and Wasmtime JIT

Self-improving code for any git repo: give it a measurable goal, a coding agent (Claude Code or Aider) proposes changes, the kernel scores them, keeps winners and rolls back failures. Open-source AlphaEvolve / FunSearch-style evolution loop, sandboxed, with an auditable ledger.

AlphaEvolve/FunSearch-style generate-evaluate-evolve bake-off: 12 LLMs evolving cache eviction policies & bin-packing heuristics on a SoC LLM cluster

An LLM rewrites your code, a fast Rust scorer grades it on held-out data, the best survives — thousands of iterations a night. Runs as a flat-fee Claude Code session, not a metered Anthropic API loop.

A fully-local engine that finds cross-domain structural bridges between research papers and evolves algorithms along them — 6 GB GPU, zero cloud calls.

NoemaEvolve extends LLM evolutionary program search with RL based reflection and a ReAct style coordination