AI-Driven Quantum Error Correction Circuit Synthesis

Can AI agents generate fault-tolerant quantum error correction circuits? This repo benchmarks LLMs and reinforcement learning on synthesizing, verifying, and optimizing stabilizer-code circuits for state preparation and syndrome extraction. All circuits use Stim format and are validated via stabilizer-based oracles.

See the research poster for details.

Research Questions

#	Question	Metric
RQ1	Can an agent generate stabilizer circuits reliably?	% stabilizer preservation
RQ2	Can an agent make a circuit fault-tolerant?	Median FT score
RQ3	Can an agent optimize without breaking FT?	Circuit volume
RQ4	Does training/fine-tuning an LLM improve results?	Same as above

Structure

Directory	Purpose
data/	Benchmarks, datasets, and LLM evaluation results
tools/	Copilot agent, MCP verification server, prompts
reinforcement_learning/	Two-agent RL system (generator + FT enforcer)
RL/	Gymnasium env for step-by-step circuit building
Examples/	Example FT circuits and verification scripts
ai_ft_prep_instructions/	Reference FT state-prep data

Setup

pip install -r tools/requirements.txt
pip install -r RL/requirements.txt

For the Copilot agent, see tools/agent-readme.md. For dataset format, see data/DATASET_FORMAT.md.