CRISP-TAI — Technical Artefact for Practical Adoption

This repository provides a technical artefact that complements the CRISP-TAI process model presented in the associated research article.
Its purpose is to support practical adoption, instantiation, and reuse of CRISP-TAI in real-world AI engineering projects.

Important scope note
This repository does not reproduce the paper, its figures, or its conceptual explanations.
The article remains the authoritative source for the motivation, theoretical foundations, and validation of CRISP-TAI.
This repository focuses exclusively on practical engineering artefacts, templates, and example instantiations that cannot be fully detailed in an academic paper.

---

What this repository is

• A technical companion to the CRISP-TAI article
• A collection of engineering artefacts that operationalise the process model
• A set of templates and examples to help practitioners apply CRISP-TAI
• A reference for traceability, evaluability, and governance-ready lifecycle documentation

---

What this repository is not

• ❌ Not a certification framework
• ❌ Not an automated compliance or auditing tool
• ❌ Not a checklist for Trustworthy AI
• ❌ Not a metrics or benchmarking library
• ❌ Not a replacement for human governance or organisational decision-making

CRISP-TAI supports evaluability and traceability, not automated judgement or normative enforcement.

---

Repository structure

/docs        Practical guidance for applying CRISP-TAI
/templates   Reusable engineering templates (CR / FR artefacts)
/examples    Cross-domain instantiations and execution traces
/logs        Illustrative rule-usage and refinement logs
/appendix    Extended tables and detailed traces referenced by examples

---

/docs — Practitioner-oriented guidance

This directory contains hands-on documentation aimed at engineers and teams applying CRISP-TAI in practice.

Typical contents include:

• getting-started.md
  How to start applying CRISP-TAI in an AI project

• lifecycle-walkthrough.md
  A phase-by-phase walkthrough showing how trustworthiness considerations evolve across the lifecycle

• rule-application-guide.md
  How Contextual Refinement (CR) and Functional Refinement (FR) rules are used in practice

• governance-notes.md
  Practical notes on roles, responsibilities, reviews, and organisational integration

These documents extend the paper by focusing on how CRISP-TAI is applied, not why it is valid.

---

/templates — Core engineering artefacts

This directory contains reusable templates that implement the CRISP-TAI refinement mechanisms.

Contextual Refinement templates

Used to support reflexive and governance-oriented reasoning:

• context-description.md
• responsibility-map.csv
• principle-prioritisation.md
• assumption-log.md

Functional Refinement templates

Used to translate refined interpretations into concrete artefacts:

• artefact-derivation.md
• traceability-matrix.csv
• monitoring-plan.md
• feedback-log.md

Phase-oriented checklists

Optional lightweight guidance per lifecycle phase:

• business-understanding.md
• data-understanding.md
• data-preparation.md
• modeling.md
• evaluation.md
• deployment-monitoring.md

All templates are technology-agnostic, domain-independent, and intended to be adapted to local engineering practices.

---

/examples — Execution-based instantiations

This directory contains illustrative instantiations of CRISP-TAI in heterogeneous settings, reflecting the execution-based validation discussed in the paper.

Examples may include:

• Retrieval-Augmented Generation in Data Spaces
• Federated Learning environments
• AI-enabled Internet of Things (TrustAIoT)

Each example typically includes:

• A short contextual description (README.md)
• An instantiation-log.md documenting applied CR/FR rules
• Selected artefacts derived during execution (anonymised or synthetic)
• Traceability and responsibility records

These examples are not reference implementations, but evidence of instantiability and practical use.

---

/logs — Refinement and traceability records

This directory provides illustrative logs showing how refinement rules may be activated and revisited over time, for example:

• Context revision triggers
• Artefact updates
• Monitoring feedback loops

These logs support evaluability and audit-readiness without prescribing auditing procedures.

---

/appendix — Supplementary material

Contains extended tables, detailed traces, or supporting material referenced by examples, when such detail would clutter the main repository structure.

---

Relationship to the research article

The CRISP-TAI article presents:

• The motivation and problem framing
• The design rationale of the process model
• The distinction between operational and reflexive reasoning
• The execution-based validation across domains

This repository provides:

• The engineering artefacts that make CRISP-TAI usable in practice
• The templates and traces that operationalise the model
• Additional material that supports adoption but exceeds article length constraints

For conceptual, methodological, and validation details, refer to the article.

---

License and reuse

This repository is intended to support reuse and adaptation in both research and industrial contexts.
See LICENSE for terms of use.

---

Citation

If you use CRISP-TAI or artefacts from this repository in academic work, please cite the associated article.

Citation details are provided in CITATION.cff.