HUMAN-Framework-Toolkit

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

The HUMAN Framework Toolkit is an open-source research resource designed to support the transparent, reflexive, and human-led use of generative artificial intelligence in psychological thematic analysis. It accompanies the HUMAN Framework (Human-led Understanding and Meaning in AI-assisted Analysis), a seven-stage methodological framework developed to guide researchers in integrating artificial intelligence into qualitative psychological research while preserving human interpretive authority, contextual sensitivity, reflexivity, and methodological transparency. The toolkit is intended for researchers, students, and practitioners who wish to use large language models and other generative AI systems as supportive analytic tools rather than as autonomous qualitative analysts. Its central principle is that AI-generated outputs should be treated as provisional analytic suggestions that require verification against the original data and critical interpretation by the researcher. The framework therefore places particular emphasis on researcher familiarisation, source-level verification, reflexive decision-making, documentation of accepted, modified, and rejected AI suggestions, and transparent reporting of how AI influenced the analytic process. This repository includes a vendor-neutral HUMAN prompt scaffold that can be adapted to different psychological research questions, epistemological positions, theoretical orientations, and levels of thematic interpretation. It also provides implementation guidance and platform-specific recommendations for major contemporary AI systems, including ChatGPT, Claude, Gemini, Microsoft Copilot, Perplexity, Grok, Mistral, DeepSeek, and Gemini Notebook/NotebookLM. These resources are designed to help researchers maintain a consistent methodological workflow while accounting for differences in platform functionality, privacy controls, source grounding, file handling, and data-governance requirements. The toolkit additionally contains templates for documenting AI-assisted analysis, including interaction logs, human decision records, coding and source-verification structures, transparency checklists, and reporting materials. A synthetic worked example demonstrates how the HUMAN Framework can be applied step by step in a psychological thematic analysis, including data preparation, researcher familiarisation, AI contextualisation, preliminary AI-assisted coding, human review, theme development, audit-trail documentation, and final disclosure of AI use. The example is entirely synthetic and is intended solely for methodological demonstration and training. An offline-first Python command-line utility is also included to support reproducible implementation of the framework. The software can initialise a study workspace, conduct a documentation preflight, generate a study-specific HUMAN prompt, create and validate AI interaction logs, and produce a draft AI-use disclosure. The software does not perform autonomous thematic analysis and does not generate final themes or interpretations. It is deliberately designed to preserve human analytic responsibility and to support documentation rather than automate qualitative judgement. The toolkit also includes AI-agent instruction files and repository guidance for development environments such as OpenAI Codex, Claude Code, Gemini CLI, GitHub Copilot, and Cursor. These files help ensure that future extensions to the software remain consistent with the methodological principles of the HUMAN Framework, including human interpretive authority, source verification, reflexivity, privacy awareness, and avoidance of automated final-theme generation. The resource is intended to evolve alongside developments in generative AI and qualitative research methodology. The GitHub repository serves as the actively maintained version of the toolkit, while this Zenodo record provides a persistent, citable archive of the released version. Researchers using the toolkit remain responsible for complying with applicable ethical approvals, informed-consent requirements, institutional policies, data-protection legislation, and platform-specific terms of use. In particular, technical compatibility with the HUMAN prompt should not be interpreted as permission to upload identifiable, clinical, or otherwise sensitive psychological data to a particular AI service. Where sensitive data are involved, researchers should use institutionally approved environments, minimise and de-identify data, document retention and processing conditions, and consult relevant data-protection and ethics procedures. The HUMAN Framework Toolkit was developed by Ricardo Tejeiro (Liverpool John Moores University) and Alberto Paramio (University of Huelva). Project repository: https://github.com/alberto-paramio/HUMAN-Framework-Toolkit Archived release: https://doi.org/10.5281/zenodo.22081358

Links and sources

Comments