qpipe-agentic-quantum-code-gen

name	qpipe-agentic-quantum-code-gen
description	LLM-based multi-agent architecture for autonomous quantum application generation from natural language requirements. Use when building agentic systems for quantum software engineering, automated quantum code generation, NL-to-quantum workflows, or quantum test optimization pipelines. Activation: qpipe, agentic quantum code generation, LLM quantum application, natural language quantum workflow, quantum test optimization agent, multi-agent quantum compilation, autonomous quantum code review.
metadata	{"arxiv_id":"2607.00939","published":"2026-07-01","tags":["quantum","software-engineering","llm","multi-agent","code-generation"]}

QPipe Agentic Quantum Code Generation

Description

LLM-based multi-agent architecture that autonomously turns natural language requirements into traceable quantum-application workflows through specialized agents.

Core Architecture

Agent Decomposition

• Requirement Parsing Agent: Extracts quantum-specific requirements from NL
• Formulation Agent: Translates requirements into quantum circuit specifications
• Code Generation Agent: Generates executable quantum code (Qiskit/Cirq/etc.)
• Review Agent: Validates quantum code correctness and optimization
• Execution Agent: Runs quantum circuits on simulators/hardware
• Verification Agent: Validates results against expected outcomes

Performance Metrics

• 100% code compilation rate across 20 benchmarks
• 96.7% application execution success rate
• Average generation: 260 seconds, 1.89M tokens per requirement
• Generated solutions outperform offline genetic algorithm baseline

Key Ablation Findings

Success depends on retaining:

1. Code-generation skills
2. Task knowledge
3. Review feedback
4. Multi-agent decomposition

When to Use

• Building agentic systems for quantum software engineering
• Automated quantum code generation from specifications
• Quantum test optimization pipelines
• NL-to-quantum workflow automation

Pitfalls

• Token costs are significant (~1.89M per requirement)
• Multi-agent coordination overhead is non-trivial
• Requires strong quantum domain knowledge in LLM prompts
• Review feedback loop is critical — skipping it degrades quality