Access datasets from Kamiwaza AIR research to support your agentic AI research and development.

---

## How Much Do LLMs Hallucinate in Document Q&A Scenarios?

Raw model outputs from the 172-billion-token hallucination study across 35 models, three context lengths, four temperatures, and three hardware platforms. Ground truth, document corpora, and test sets provided separately.

📦 **[Download RIKER2_March2026.zip](https://research.kamiwaza.ai/HowMuchDoLLMsHallucinateInDocQA/RIKER2_March2026.zip)** (4.34 GB) — Raw model outputs

📦 **[Download RIKER2_corpora_groundtruth_testsets.zip](https://research.kamiwaza.ai/HowMuchDoLLMsHallucinateInDocQA/RIKER2_corpora_groundtruth_testsets.zip)** (0.60 MB) — Ground truth, corpora, and test sets

🔗 **[Read the paper](https://docs.kamiwaza.ai/research/papers/riker2)** | [arXiv](https://arxiv.org/abs/2603.08274)

**Citation:**

```bibtex
@article{roig2026hallucinate,

title={How Much Do LLMs Hallucinate in Document Q\&A Scenarios? A 172-Billion-Token Study Across Temperatures, Context Lengths, and Hardware Platforms},

author={Roig, JV},

year={2026},

eprint={2603.08274},

archivePrefix={arXiv},

primaryClass={cs.CL},

url={https://arxiv.org/abs/2603.08274}

}
```

---

## Scalable and Reliable Evaluation of AI Knowledge Retrieval Systems (aka "RIKER")

Generated ground truth, document corpora, test sets, and the various model raw results.

📦 **[Download riker_2025.zip](https://research.kamiwaza.ai/RIKER_LLM_2025/riker_2025.zip)** (612.45 MB)

🔗 **[Read the paper](https://docs.kamiwaza.ai/research/papers/riker)**

**Citation:**

```bibtex
@techreport{roig2025riker,

title={Scalable and Reliable Evaluation of AI Knowledge Retrieval Systems: RIKER and the Coherent Simulated Universe},

author={Roig, JV},

institution={Kamiwaza AI},

year={2025},

url={https://docs.kamiwaza.ai/research/papers/riker}

}
```

---

## Towards a Standard, Enterprise-Relevant Agentic AI Benchmark (aka "KAMI v0.1")

Test suite definitions and evaluation code from our enterprise-focused agentic AI benchmark study.

📦 **[Download KAMI_v0.1_2025-12-17.zip](https://research.kamiwaza.ai/KAMI_v0.1/KAMI_v0.1_2025-12-17.zip)** (999.62 MB)

🔗 **[Read the paper](https://docs.kamiwaza.ai/research/papers/kami-v0-1)**

**Citation:**

```bibtex
@techreport{roig2025kami,

title={Towards a Standard, Enterprise-Relevant Agentic AI Benchmark: Lessons from 5.5 billion tokens' worth of agentic AI evaluations},

author={Roig, JV},

institution={Kamiwaza AI},

year={2025},

url={https://docs.kamiwaza.ai/research/papers/kami-v0-1}

}
```

---

## How Do LLMs Fail In Agentic Scenarios?

Execution traces from our qualitative analysis of LLM failure modes in agentic simulations.

📦 **[Download HowDoLLMsFail_2025.zip](https://research.kamiwaza.ai/HowDoLLMsFail_2025/HowDoLLMsFail_2025.zip)** (2.60 MB)

🔗 **[Read the paper](https://docs.kamiwaza.ai/research/papers/llm-agentic-failures)**

**Citation:**

```bibtex
@techreport{roig2025llmfailures,

title={How Do LLMs Fail In Agentic Scenarios? A Qualitative Analysis of Success and Failure Scenarios of Various LLMs in Agentic Simulations},

author={Roig, JV},

institution={Kamiwaza AI},

year={2025},

url={https://docs.kamiwaza.ai/research/papers/llm-agentic-failures}

}
```