Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

Quantum AI Reservoirs with Optional Graph Context across LLM, RAG and Agentic Workflows

Maksym Husarov

Published Oct 3, 2026• 0 citations
Publisher

Abstract

This study evaluates an external simulator-backed quantum/classical guidance component with optional graph context across direct LLM advice, retrieval-augmented generation (RAG), and tool-using Agentic workflows. Three conditions are compared: Baseline, QR only, and QR + graph. The RAG and Agentic evaluation comprises 17 tasks, three conditions, and three repeats per workflow, producing 306 answers. A deterministic evaluator measures complete success through structured decision correctness, quotation validity, decisive-source coverage, and, for Agentic workflows, consistency with a simulated commit. Complete RAG success was 88.2% for Baseline, 92.2% for QR only, and 98.0% for QR + graph. Complete Agentic success was 62.7%, 54.9%, and 86.3%, respectively. Relative to QR only, graph context increased success by 5.9 percentage points in RAG and 31.4 points in Agentic. The combined configuration exceeded Agentic Baseline in all three repeats. Task-conditional bootstrap intervals support a clearer benefit in Agentic; the RAG intervals include zero. A separate single-request advisory comparison produced weighted LLM-review scores of 87.5, 95.0, and 95.0 out of 100. These exploratory scores are reported separately and are not accuracy percentages. The release includes the article and two public JSON supplements containing anonymized trial outcomes, summaries, evaluation settings, and advisory score vectors. Proprietary implementation, prompts, source passages, and model response texts are withheld. The findings concern the integrated guidance and evidence-acquisition system on the evaluated tasks. They do not establish quantum computational advantage, isolate the quantum component as the cause of improvement, or demonstrate general superiority. The main evaluation is a follow-up run on previously inspected tasks, not an unseen holdout.

DOI
10.5281/zenodo.23111918

Related papers

Quantum Reservoirs as External Decision Layers for LLMs: Evidence from RAG and Agentic Systems

Maksym Husarov

I evaluated whether a quantum AI reservoir can improve the decisions produced by retrieval-augmented generation (RAG) and tool-using Agentic workflows. Two complementary experiments used the same five workflow variants: RAG, RAG + QR, Agentic, Agentic + QR, and LLM + QR. Each experiment contained 201 main tasks and three repeats, yielding 6,030 answers overall. The first experiment used synthetic planning scenarios; the second used numerical quantities extracted from quantum-domain books and technical papers. In the designated first repeat, QR increased complete structured success from 75.1% to 90.5% for synthetic RAG and from 27.4% to 52.2% for synthetic Agentic. On real-source tasks, the corresponding rates increased from 46.8% to 57.7% and from 34.3% to 55.7%. Both paired improvements were positive in every repeat of both experiments. LLM + QR achieved 94.0% on synthetic tasks and 54.7% on real-source tasks in the first repeat, exceeding both unassisted baselines in each experiment. The results support the integrated reservoir-guidance module as a useful addition to the evaluated decision workflows and motivate a simpler single-call architecture when all required evidence is available. The study evaluates a simulator-backed quantum/classical module; it does not establish quantum computational advantage or isolate the quantum component from its classical guidance components.

•Sep 20, 2026•0 citations

Quantum Reservoir-Assisted Context Orchestration for Large Language Models: A 500-Prompt Paired Evaluation

Maksym Husarov

Large language models increasingly operate within systems that provide retrieved knowledge, policies, tool descriptions, workflow state, and other contextual signals. This work investigates Quantum Reservoir Assist, a pre-generation context-orchestration layer that uses a CUDA-Q-based quantum reservoir process to rank candidate contextual signals and construct dynamic decision-support guidance for an otherwise unchanged downstream language model. We evaluate the complete pipeline using a paired benchmark of 500 production-like prompts across ten categories, including architecture and system design, troubleshooting, multi-constraint reasoning, incident response, agent tool planning, security and reliability, and technical explanation. Both experimental paths use the same downstream LLM and generation settings. Across the benchmark, Quantum Reservoir Assist produced a mean absolute increase of approximately 0.172 in the system’s internal answer-readiness/accuracy proxy. The strongest mean proxy improvements occurred in troubleshooting, ambiguous or underspecified requests, conflicting constraints, security and reliability, and multi-constraint tasks. Reservoir assistance also introduced measurable operational overhead, including increased downstream token usage and substantial preprocessing latency under the CUDA-Q simulation configuration used in this experiment. The results demonstrate systematic effects of quantum-reservoir-assisted pre-generation context orchestration, but do not establish ground-truth answer-accuracy improvement or quantum computational advantage. The study therefore motivates controlled ablations against static guidance and classical context-ranking methods.

•Aug 26, 2026•0 citations

Revisiting Quantum Code Generation: Where Should Domain Knowledge Live?

Oscar Novo, Oscar Bastidas-Jossa, Alberto Calvo +2 more

Recent advances in large language models (LLMs) have enabled the automation of an increasing number of programming tasks, including code generation for scientific and engineering domains. In rapidly evolving software ecosystems such as quantum software development, where frameworks expose complex abstractions, a central question is how best to incorporate domain knowledge into LLM-based assistants while preserving maintainability as libraries evolve. In this work, we study specialization strategies for Qiskit code generation using the Qiskit-HumanEval benchmark. We compare a parameter-specialized fine-tuned baseline introduced in prior work against a range of recent general-purpose LLMs enhanced with retrieval-augmented generation (RAG) and agent-based inference with execution feedback. Our results show that modern general-purpose LLMs consistently outperform the parameter-specialized baseline. While the fine-tuned model achieves approximately 47% pass@1 on Qiskit-HumanEval, recent general-purpose models reach 60-65% under zero-shot and retrieval-augmented settings, and up to 85% for the strongest evaluated model when combined with iterative execution-feedback agents -representing an improvement of more than 20% over zero-shot general-purpose performance and more than 35% over the parameter-specialized baseline. Agentic execution feedback yields the most consistent improvements, albeit at increased runtime cost, while RAG provides modest and model-dependent gains. These findings indicate that performance gains can be achieved without domain-specific fine-tuning, instead relying on inference-time augmentation, thereby enabling a more flexible and maintainable approach to LLM-assisted quantum software development.

cs.LGCross-listedarXiv:2603.22184•Mar 24, 2026•0 citations

Quantum-Enhanced Retrieval-Augmented Generation for Hallucination Reduction in Large Language Models

Praveenkumar Seepana

Although the performance of LLMs on a wide variety of natural language processing problems demonstrates remarkable ability, hallucinated responses are introduced as one of the key weaknesses of LLMs, especially in knowledge-intensive applications, where fidelity to facts is paramount. While effective in reducing hallucinations, the context retrieved during the RAG operations is still very often suboptimal with respect to the external documents used in the retrieval stage, and being based on cosine similarity and nearest-neighbour search, these models typically do not return optimal context to support factual generation. The authors propose a five-stage solution, called Quantum-Enhanced Retrieval-Augmented Generation (QERAG), which combines the use of a quantum-inspired probability amplitude document ranking, a Context Utility Score (CUS) optimisation engine, and a Hallucination Verification Agent (HVA) based on a formal Hallucination Reduction Index (HRI). The Quantum Relevance Score (QRS) uses interference based, Amplitude Encoding on candidate Document Sets to yield a normalised relevance distribution that emphasises the relatively higher value generated in the discriminative context as compared to other locally similar passages in a document set. QERAG outperforms Standard LLM and Traditional RAG models on all five Natural benchmark across response accuracy, with a score of 92.8% vs. 74.3%, and hallucination rate, 4.3% vs. 18.2%, respectively. They achieve a significant overall improvement by using quantum-inspired retrieval layer, which achieves a 3.5-percentage-point reduction in hallucination rate when compared to cosine-similarity RAG, with this single layer showing more reduction than the rest of the pipeline layers. The results of an ablation study show the individual contribution of every pipeline stage, and the quantum-inspired retrieval layer exhibits the greatest reduction of a 3.5 percentage points over cosine-similarity RAG.

International Journal of Computational Science and Engineering Research•Aug 3, 2026•0 citations

Benchmarking the Benchmarks for Quantum Code Generation with LLMs

Yunhe Feng, Xiaokun Yang, Kewei Sha +1 more

Large Language Models (LLMs) are increasingly deployed as quantum coding copilots, assisting researchers by generating complex programs for frameworks such as Qiskit, PennyLane, and OpenQASM. However, accurately evaluating the practical utility of these models remains a significant challenge. Existing benchmarks are highly fragmented, ranging from syntax-centric unit tests to simulator- and hardware-aware metrics, making results difficult to compare across different languages and hardware targets. To address this fragmentation, this paper systematically reviews recent evaluation suites and post-training methodologies for quantum code generation. We distill current evaluation criteria into four practical dimensions: syntax and API validity, semantic and physical correctness, resource and hardware feasibility, and execution-based feedback. Building upon these dimensions, we present a unified taxonomy that maps current benchmarks to emerging model adaptation techniques, specifically retrieval-augmented generation (RAG), fine-tuning, multi-agent workflows, and reinforcement learning (RL). Our analysis highlights a clear shift: moving away from prompt-only baselines and toward agentic workflows driven by quantum-verifiable rewards. This shift directly targets the semantic gap between superficially compilable code and physically executable quantum programs. In addition, we identify key research opportunities-such as standardized semantic metrics and integration with quantum error correction-aimed at fostering a more reproducible, verifiable, and hardware-grounded ecosystem for automated quantum software development.

IEEE Dallas Circuits and Systems Conference (DCAS)•Apr 10, 2026•0 citations

A Protocol-Guided LLM Agent for Quantum Program Synthesis and Execution

Ming-Kang Ho, Tai-Yue Li

This study evaluates a protocol-guided large language model (LLM) agent workflow for quantum program synthesis and execution. A versioned YAML protocol specifies interface and quantum-semantic requirements while leaving circuit design to the model. The workflow combines Qiskit circuit generation, evaluator-guided repair, and quantum processing unit (QPU) deployment. A matched ablation evaluates four LLM endpoints using the variational quantum eigensolver (VQE) for H$_2$, the quantum approximate optimization algorithm (QAOA) for MaxCut, and Fashion-MNIST quantum-kernel classification. Across 240 trials, protocol guidance increased evaluator completion from 80.0\% to 99.2\%, reduced mean repairs from 2.56 to 0.75, and reduced mean agent time from 55.70 to 30.58~s. Provider-reported token use decreased for three of four endpoints. Completed VQE trials meeting chemical accuracy increased from 16/40 to 29/40, while valid QAOA approximation ratios remained comparable. Semantic review identified rank-one fidelity kernels, reinforcing the distinction between evaluator completion and semantic validity. A 210-run IBM QPU study evaluated the combined generation-and-selection pipelines and confirmed deployment feasibility. The results support protocol guidance as a software-reliability layer that must be complemented by quantum-semantic validation.

quant-pharXiv:2609.20258•Jul 29, 2026•0 citations
This study evaluates an external simulator-backed quantum/classical guidance component with optional graph context across direct LLM advice, retrieval-augmented generation (RAG), and tool-using Agentic workflows. Three conditions are compared: Baseline, QR only, and QR + graph. The RAG and Agentic evaluation comprises 17 tasks, three conditions, and three repeats per workflow, producing 306 answers. A deterministic evaluator measures complete success through structured decision correctness, quotation validity, decisive-source coverage, and, for Agentic workflows, consistency with a simulated commit. Complete RAG success was 88.2% for Baseline, 92.2% for QR only, and 98.0% for QR + graph. Complete Agentic success was 62.7%, 54.9%, and 86.3%, respectively. Relative to QR only, graph context increased success by 5.9 percentage points in RAG and 31.4 points in Agentic. The combined configuration exceeded Agentic Baseline in all three repeats. Task-conditional bootstrap intervals support a clearer benefit in Agentic; the RAG intervals include zero. A separate single-request advisory comparison produced weighted LLM-review scores of 87.5, 95.0, and 95.0 out of 100. These exploratory scores are reported separately and are not accuracy percentages. The release includes the article and two public JSON supplements containing anonymized trial outcomes, summaries, evaluation settings, and advisory score vectors. Proprietary implementation, prompts, source passages, and model response texts are withheld. The findings concern the integrated guidance and evidence-acquisition system on the evaluated tasks. They do not establish quantum computational advantage, isolate the quantum component as the cause of improvement, or demonstrate general superiority. The main evaluation is a follow-up run on previously inspected tasks, not an unseen holdout.
I evaluated whether a quantum AI reservoir can improve the decisions produced by retrieval-augmented generation (RAG) and tool-using Agentic workflows. Two complementary experiments used the same five workflow variants: RAG, RAG + QR, Agentic, Agentic + QR, and LLM + QR. Each experiment contained 201 main tasks and three repeats, yielding 6,030 answers overall. The first experiment used synthetic planning scenarios; the second used numerical quantities extracted from quantum-domain books and technical papers. In the designated first repeat, QR increased complete structured success from 75.1% to 90.5% for synthetic RAG and from 27.4% to 52.2% for synthetic Agentic. On real-source tasks, the corresponding rates increased from 46.8% to 57.7% and from 34.3% to 55.7%. Both paired improvements were positive in every repeat of both experiments. LLM + QR achieved 94.0% on synthetic tasks and 54.7% on real-source tasks in the first repeat, exceeding both unassisted baselines in each experiment. The results support the integrated reservoir-guidance module as a useful addition to the evaluated decision workflows and motivate a simpler single-call architecture when all required evidence is available. The study evaluates a simulator-backed quantum/classical module; it does not establish quantum computational advantage or isolate the quantum component from its classical guidance components.
Large language models increasingly operate within systems that provide retrieved knowledge, policies, tool descriptions, workflow state, and other contextual signals. This work investigates Quantum Reservoir Assist, a pre-generation context-orchestration layer that uses a CUDA-Q-based quantum reservoir process to rank candidate contextual signals and construct dynamic decision-support guidance for an otherwise unchanged downstream language model. We evaluate the complete pipeline using a paired benchmark of 500 production-like prompts across ten categories, including architecture and system design, troubleshooting, multi-constraint reasoning, incident response, agent tool planning, security and reliability, and technical explanation. Both experimental paths use the same downstream LLM and generation settings. Across the benchmark, Quantum Reservoir Assist produced a mean absolute increase of approximately 0.172 in the system’s internal answer-readiness/accuracy proxy. The strongest mean proxy improvements occurred in troubleshooting, ambiguous or underspecified requests, conflicting constraints, security and reliability, and multi-constraint tasks. Reservoir assistance also introduced measurable operational overhead, including increased downstream token usage and substantial preprocessing latency under the CUDA-Q simulation configuration used in this experiment. The results demonstrate systematic effects of quantum-reservoir-assisted pre-generation context orchestration, but do not establish ground-truth answer-accuracy improvement or quantum computational advantage. The study therefore motivates controlled ablations against static guidance and classical context-ranking methods.
Recent advances in large language models (LLMs) have enabled the automation of an increasing number of programming tasks, including code generation for scientific and engineering domains. In rapidly evolving software ecosystems such as quantum software development, where frameworks expose complex abstractions, a central question is how best to incorporate domain knowledge into LLM-based assistants while preserving maintainability as libraries evolve. In this work, we study specialization strategies for Qiskit code generation using the Qiskit-HumanEval benchmark. We compare a parameter-specialized fine-tuned baseline introduced in prior work against a range of recent general-purpose LLMs enhanced with retrieval-augmented generation (RAG) and agent-based inference with execution feedback. Our results show that modern general-purpose LLMs consistently outperform the parameter-specialized baseline. While the fine-tuned model achieves approximately 47% pass@1 on Qiskit-HumanEval, recent general-purpose models reach 60-65% under zero-shot and retrieval-augmented settings, and up to 85% for the strongest evaluated model when combined with iterative execution-feedback agents -representing an improvement of more than 20% over zero-shot general-purpose performance and more than 35% over the parameter-specialized baseline. Agentic execution feedback yields the most consistent improvements, albeit at increased runtime cost, while RAG provides modest and model-dependent gains. These findings indicate that performance gains can be achieved without domain-specific fine-tuning, instead relying on inference-time augmentation, thereby enabling a more flexible and maintainable approach to LLM-assisted quantum software development.
Although the performance of LLMs on a wide variety of natural language processing problems demonstrates remarkable ability, hallucinated responses are introduced as one of the key weaknesses of LLMs, especially in knowledge-intensive applications, where fidelity to facts is paramount. While effective in reducing hallucinations, the context retrieved during the RAG operations is still very often suboptimal with respect to the external documents used in the retrieval stage, and being based on cosine similarity and nearest-neighbour search, these models typically do not return optimal context to support factual generation. The authors propose a five-stage solution, called Quantum-Enhanced Retrieval-Augmented Generation (QERAG), which combines the use of a quantum-inspired probability amplitude document ranking, a Context Utility Score (CUS) optimisation engine, and a Hallucination Verification Agent (HVA) based on a formal Hallucination Reduction Index (HRI). The Quantum Relevance Score (QRS) uses interference based, Amplitude Encoding on candidate Document Sets to yield a normalised relevance distribution that emphasises the relatively higher value generated in the discriminative context as compared to other locally similar passages in a document set. QERAG outperforms Standard LLM and Traditional RAG models on all five Natural benchmark across response accuracy, with a score of 92.8% vs. 74.3%, and hallucination rate, 4.3% vs. 18.2%, respectively. They achieve a significant overall improvement by using quantum-inspired retrieval layer, which achieves a 3.5-percentage-point reduction in hallucination rate when compared to cosine-similarity RAG, with this single layer showing more reduction than the rest of the pipeline layers. The results of an ablation study show the individual contribution of every pipeline stage, and the quantum-inspired retrieval layer exhibits the greatest reduction of a 3.5 percentage points over cosine-similarity RAG.
Large Language Models (LLMs) are increasingly deployed as quantum coding copilots, assisting researchers by generating complex programs for frameworks such as Qiskit, PennyLane, and OpenQASM. However, accurately evaluating the practical utility of these models remains a significant challenge. Existing benchmarks are highly fragmented, ranging from syntax-centric unit tests to simulator- and hardware-aware metrics, making results difficult to compare across different languages and hardware targets. To address this fragmentation, this paper systematically reviews recent evaluation suites and post-training methodologies for quantum code generation. We distill current evaluation criteria into four practical dimensions: syntax and API validity, semantic and physical correctness, resource and hardware feasibility, and execution-based feedback. Building upon these dimensions, we present a unified taxonomy that maps current benchmarks to emerging model adaptation techniques, specifically retrieval-augmented generation (RAG), fine-tuning, multi-agent workflows, and reinforcement learning (RL). Our analysis highlights a clear shift: moving away from prompt-only baselines and toward agentic workflows driven by quantum-verifiable rewards. This shift directly targets the semantic gap between superficially compilable code and physically executable quantum programs. In addition, we identify key research opportunities-such as standardized semantic metrics and integration with quantum error correction-aimed at fostering a more reproducible, verifiable, and hardware-grounded ecosystem for automated quantum software development.
This study evaluates a protocol-guided large language model (LLM) agent workflow for quantum program synthesis and execution. A versioned YAML protocol specifies interface and quantum-semantic requirements while leaving circuit design to the model. The workflow combines Qiskit circuit generation, evaluator-guided repair, and quantum processing unit (QPU) deployment. A matched ablation evaluates four LLM endpoints using the variational quantum eigensolver (VQE) for H2_22​, the quantum approximate optimization algorithm (QAOA) for MaxCut, and Fashion-MNIST quantum-kernel classification. Across 240 trials, protocol guidance increased evaluator completion from 80.0\% to 99.2\%, reduced mean repairs from 2.56 to 0.75, and reduced mean agent time from 55.70 to 30.58~s. Provider-reported token use decreased for three of four endpoints. Completed VQE trials meeting chemical accuracy increased from 16/40 to 29/40, while valid QAOA approximation ratios remained comparable. Semantic review identified rank-one fidelity kernels, reinforcing the distinction between evaluator completion and semantic validity. A 210-run IBM QPU study evaluated the combined generation-and-selection pipelines and confirmed deployment feasibility. The results support protocol guidance as a software-reliability layer that must be complemented by quantum-semantic validation.