Labsco
JDerekLomas logo

Quantum Vibecoding

from JDerekLomas

Run real quantum experiments on real quantum hardware via natural language. Three MCP servers in one PyPI package: Quantum Inspire Tuna-9 (superconducting qubits), IBM Quantum, and quantum RNG. One-line install: claude mcp add qi-circuits -- uvx --from "quantum-vibecoding-mcp[qi]" qvc-qi

🔥🔥🔥🔥✓ VerifiedAccount requiredAdvanced setup
#documents#git

AI x Quantum — TU Delft / Quantum Inspire

How might generative AI accelerate quantum computing?

An open research initiative exploring AI-driven quantum computing research. We build autonomous agents that run quantum experiments across multiple hardware backends, replicate published papers, and benchmark LLM capabilities on quantum tasks.

Live site: https://quantuminspire.vercel.app

Current results

Experiments (22 results across 7 study types)

StudyBackendsKey result
Bell State CalibrationEmulator, IBM Marrakesh, IBM Torino, Tuna-9100% / 99.05% / varies fidelity
GHZ State (3q)Emulator, IBM Marrakesh, IBM Torino, Tuna-9100% / 98.14% fidelity
H2 VQE (2q)Emulator, IBM Marrakesh, IBM Torino, Tuna-9-1.1385 Ha emulator (chemical accuracy)
QRNG CertificationTuna-9 raw + debiased, EmulatorRaw fails NIST; debiased passes all
Randomized BenchmarkingEmulator99.95% gate fidelity
QAOA MaxCutEmulator87% approximation ratio
Quantum VolumeEmulator + Tuna-9QV 16 (4q pass, 8/10 circuits)

Additional hardware experiments: connectivity probe (Tuna-9 topology), repetition code (3q QEC), detection code (emulator).

Paper replications (3 papers, 13 claims)

PaperClaims testedPass rate
Sagastizabal 2019 (H2 VQE)743% (emulator pass, hardware fail)
Peruzzo 2014 (HeH+ VQE)3100% (emulator)
Cross 2019 (Quantum Volume)3100% (emulator)

Hardware access

BackendQubitsAccess
QI Emulator (qxelarator)ConfigurableLocal, no auth needed
QI Tuna-99 (6 usable)QI member 2108
IBM Marrakesh156IBM Quantum (free tier, 10 min/month)
IBM Torino133IBM Quantum
IBM Fez156IBM Quantum

Architecture

Website (Next.js 14 + Tailwind + Three.js)

RouteDescription
/Research home — hero, experiments overview, agent architecture
/experimentsExperiment dashboard — grouped by type, backend badges
/experiments/[id]Study detail — abstract, research question, results, visualizations
/replicationsPaper replication dashboard — claims vs measured, cross-backend
/blogResearch blog (7 posts)
/learnInteractive quantum learning page
/bloch-sphere, /state-vector, etc.Interactive quantum visualizations

Agents (agents/)

AgentPurpose
orchestrator.pyPipeline coordinator
experiment_daemon.pyQueue -> submit -> analyze -> store results
benchmark_agent.pyLLM benchmark runner
replication_agent.pyPaper registry + run/analyze replications
replication_analyzer.pyCompare results vs published claims
qec_decoder.pyQuantum error correction decoder

MCP servers (mcp-servers/)

ServerPurpose
qi-circuitsSubmit/check circuits on Quantum Inspire hardware
qrngQuantum random number generation
ibm-quantumIBM Quantum hardware access

Experiment result JSON schema (v1.0)

All result files in experiments/results/ follow this schema:

{
  "schema_version": "1.0",
  "id": "bell-calibration-001-ibm",
  "type": "bell_calibration",
  "backend": "ibm_marrakesh",
  "backend_qubits": 156,
  "job_id": "d65kqpoqbmes739d1k2g",
  "submitted": "2026-02-10T15:24:38Z",
  "completed": "2026-02-10T15:24:38Z",
  "parameters": { "shots": 4096 },
  "raw_counts": { ... },
  "analysis": { ... },
  "circuit_cqasm": "version 3.0\n...",
  "errors": null
}
  • schema_version: always "1.0"
  • backend_qubits: qubit count of backend (null for emulators)
  • job_id: hardware job ID (null for emulator/local runs)

Stack

  • Quantum: Qiskit 2.1, PennyLane 0.44, QI SDK 3.5.1, OpenFermion, PySCF
  • AI: Claude, Gemini, GPT (via respective APIs)
  • Web: Next.js 14, Tailwind, Three.js
  • Hardware: Quantum Inspire Tuna-9 (9q), IBM Marrakesh (156q), IBM Torino (133q), IBM Fez (156q)
  • Python: 3.9-3.13 (3.14 breaks qxelarator)