Metadata Harvesting, AI Processing & Copyright Policy
1. Purpose
Mtafiti is a scholarly discovery and research assistant that enables users to search, explore, and interact with research outputs from participating institutional repositories using artificial intelligence.
The platform enhances discovery by:
- Indexing descriptive metadata exposed by repositories through public APIs (OAI-PMH, DSpace REST, and equivalent standards).
- Generating semantic embeddings of metadata and abstracts to improve search relevance.
- Providing an optional conversational interface where users may ask questions grounded in the metadata and abstracts of documents they explicitly select.
Mtafiti is not a repository. It does not store, host, redistribute, or provide direct access to full-text copyrighted publications. All access to full content is directed to the authoritative source repository.
2. Scope
This policy applies to:
- All metadata harvested from participating institutional repositories.
- All AI processing (embedding, indexing, reranking, summarization) performed on that metadata.
- All user-facing features including search, expert discovery, document summarization, and conversational chat.
- All individuals and systems operating, maintaining, or using the Mtafiti platform.
3. Metadata Harvesting Principles
Mtafiti harvests only information that repositories intentionally expose for public discovery through standardized, machine-readable interfaces.
3.1 Metadata That May Be Harvested
- Title
- Author(s) and affiliations
- Abstract or description
- Keywords and subject classifications
- Publication date and year
- Institution / repository
- Collection and community
- Handle, DOI, or persistent URI
- Language and document type
- Licensing or rights statements (as published)
3.2 Content That Shall NOT Be Harvested or Stored
- Full-text PDF files
- Supplementary copyrighted files (datasets, appendices, media)
- Images or figures extracted from publications
- Restricted, embargoed, or access-controlled content
- Content behind authentication or paywalls
3.3 On-Demand Processing of User-Selected Documents
When a user explicitly selects a document and requests AI-assisted interaction (e.g., summarization or chat):
- Mtafiti retrieves the document only from the authoritative repository via its public access link.
- The document is processed transiently in memory for the sole purpose of the user's session.
- Only the derived embeddings of the document's content are retained for the duration of the conversation, in accordance with Section 5.
- Raw full-text files are not permanently stored, redistributed, or made available for download through Mtafiti.
This aligns with fair-dealing principles under the Copyright Act of Kenya (Cap. 130) for research and private study.
4. Copyright Statement
Copyright of all publications remains with the respective authors, publishers, or owning institutions.
Mtafiti shall:
- Not claim ownership of any harvested metadata or derived embeddings.
- Not redistribute, resell, or sublicense copyrighted full-text documents.
- Display clear, direct links to the original repository record for every search result.
- Respect embargoes, access restrictions, and rights statements defined by source repositories.
- Honor takedown or removal requests from rights holders and repositories (see Section 7).
Rights holders and repositories retain full control over their content and may request exclusion from the index at any time.
5. AI Processing
Mtafiti uses artificial intelligence solely to improve information discovery and research assistance.
5.1 AI Processing May Include
- Semantic indexing of metadata and abstracts
- Natural-language and cross-lingual search
- Similar-publication recommendations
- Topic clustering and knowledge mapping
- Document summarization (from metadata and abstracts)
- Grounded question-answering within user-selected documents
5.2 Nature of Embeddings
Embeddings are numerical, derived representations of text. They:
- Do not reproduce the original text in human-readable form.
- Are not substitutes for the original publication.
- Are used exclusively for search, ranking, and retrieval.
- Are stored in a secure vector database with access controls.
5.3 Model Transparency
Mtafiti documents the AI models and versions used for embedding, reranking, and generation, and updates them in line with security and quality reviews.
6. Attribution
Every search result and AI-generated response shall display, where available:
- Publication title
- Author(s)
- Source institution / repository
- Year of publication
- Handle, DOI, or persistent identifier
- Direct link to the original repository record
- A visible "View at source" action
The source repository shall always be acknowledged as the authoritative record. Mtafiti explicitly states that its outputs are derived summaries and not substitutes for the original work.
7. Respect for Repository Rights
Before and during harvesting, Mtafiti shall:
- Read the repository's OAI-PMH Identify record or equivalent capability statement.
- Respect repository-level rights statements, licenses, and access policies.
- Exclude collections marked as restricted, embargoed, or withdrawn.
- Update records incrementally to reflect corrections, retractions, or removals.
- Honor opt-out requests from repositories or rights holders.
7.1 Takedown & Removal
If a participating institution or rights holder requests removal of metadata from the index, Mtafiti shall:
- Acknowledge the request within 5 working days.
- Remove the affected records within 14 working days.
- Confirm removal in writing to the requesting party.
- Update the incremental state so records are not re-harvested.
8. Data Protection
Mtafiti processes personal information in accordance with the Data Protection Act, 2019 (Kenya) and applicable international standards.
- The platform primarily indexes publicly available scholarly metadata.
- Mtafiti does not harvest sensitive personal information beyond publicly available academic attribution (e.g., author names, affiliations).
- Conversation logs are retained for the duration of a session and not linked to individual users unless the user is authenticated.
- Authenticated users retain the right to request deletion of their interaction data.
9. Transparency & User Trust
Mtafiti is committed to transparency:
- Search results are labeled with their source repository.
- AI-generated answers cite the specific document(s) and section(s) used.
- The platform displays a prominent disclaimer that its outputs are derived and not authoritative legal, medical, or professional advice.
- Users are directed to the original repository for verification and citation.
10. Responsibilities
| Party | Responsibility |
|---|---|
| Mtafiti Platform | Harvest metadata ethically; maintain attribution; respect rights; secure derived data; honor takedowns |
| Source Repository | Maintain accurate metadata; publish clear rights statements; communicate changes |
| Rights Holder / Author | Identify authoritative versions; communicate removal requests where needed |
| Users | Access publications through the original repository; respect copyright; cite sources |
11. Disclaimer
The Mtafiti AI Discovery Platform is a discovery and research assistance service. Metadata is harvested from publicly accessible institutional repositories. Copyright and intellectual property rights remain with the respective rights holders.
- AI outputs are derived summaries and interpretations, not authoritative reproductions.
- Access to the full publication is provided only through the originating repository.
- Mtafiti does not provide legal, medical, financial, or professional advice.
- Users are responsible for verifying information against the authoritative source.
12. Review & Governance
This policy shall be reviewed:
- Every 3 years, or
- Earlier if required by changes in:
- Kenyan legislation (Copyright Act, Data Protection Act, Access to Information Act)
- Repository standards (OAI-PMH, DSpace, OpenAIRE)
- AI governance frameworks (e.g., Kenya's National AI Strategy)
- Institutional or rights-holder feedback
13. Contact
Mtafiti Platform - Knowledge Discovery Services
Email: mtafiti@viel.co.ke
Web: https://mtafiti.viel.co.ke/policy