Archive Assistance

Document Intelligence Assistance (DIA) uses our own AI models to analyse your documents and organise your collection. It:

• Extracts key information from each document, such as people, places, organisations and dates, and adds it as searchable tags.

• Categorises documents by type and subject, so the collection can be browsed and filtered in meaningful ways.

• Clusters related documents together, bringing out connections and themes across the collection that are hard to spot by hand.

This makes your collection richer, easier to search and quicker to catalogue, and no third-party AI provider is involved.

Document Archive Assistant (DAA), also known as Archive Assistant, is an AI research assistant for your collection. When you ask a question, it works out what you are looking for, searches your archive, and summarises what it finds, with references to the exact documents and pages it used.

It works only with your archive. It does not search the internet or use outside sources to add to its answers.

DIA and DAA handle data in different ways.

DIA: All processing uses our own in-house AI models. They run on our on-premises servers or, temporarily, on dedicated GPU servers in London hosted by our partner RunPod. Your documents are never sent to a third-party AI provider, and no third party uses your data to train models.

DAA (opt-in only): When you ask a question, DAA finds the most relevant passages in your collection and sends only those text excerpts, never page images or full documents, to an external large language model (LLM) provider, which writes the answer. We use Google's enterprise Gemini service. Under Google's enterprise terms, your data is never used to train or improve their services.

Your data is stored on our own on-premises servers in the UK. Some processing runs temporarily on dedicated GPU servers in London provided by our hosting partner, RunPod. This is used for DIA only, and data is not stored there once processing is complete.

If you opt in to DAA, the text excerpts sent to answer your questions are processed by Google under its enterprise data protection terms.

Within Microform, only specifically authorised employees with a genuine business need can access customer data. This includes support staff where their role requires it. Access:

• is limited to the minimum needed for each person's duties

• must be approved by the data owner or authorised management

• is reviewed regularly and removed when employment or contract work ends

• requires multi-factor authentication for privileged access to production systems

Outside Microform, your data is only handled by our hosting partner's servers while DIA processing is running and, if you opt in to DAA, by our LLM provider as described above.

With DIA, no. DIA uses only our own AI models, which run on our servers or on our UK hosting partner's servers. No third-party AI provider is involved.

With DAA, which is opt-in, our LLM provider, Google (enterprise Gemini service), receives only the excerpts needed to answer your question. It never receives your full documents or your full collection, and it cannot use your data for training.

We use technical and organisational safeguards to protect customer data, including:

• Encryption: Data is encrypted in transit (TLS 1.2 or later) and at rest (AES-256), on both our on-premises systems and cloud services.

• Independent audits and testing: Regular third-party security assessments, ISO 27001 certification audits and surveillance visits, and penetration testing.

• Data protection and integrity: Compliance monitoring, data accuracy checks before transfer, regular backups, and annual restore tests to verify that backups can be recovered.

• Access controls: Role-based access permissions, and multi-factor authentication for access to sensitive data and critical systems.

• Certifications: ISO 27001:2022, Cyber Essentials and BS 10008:2020.

The wording may change between answers, because AI models do not produce exactly the same text every time. The answers are drawn from the same source documents and include references, so you can always check them against the originals.

When you ask a question, the assistant first works out what you are asking. It then runs targeted searches across your collection's documents and page text, and writes an answer using only the passages it finds, with references that link back to the source pages.

The assistant does not search the internet or any other outside source. Its answers are based on what it finds in your archive.

Like all AI tools, the assistant can sometimes make mistakes. Please check important information against the referenced documents.

Yes. You keep full ownership of your material. This includes any metadata, tags and AI-generated outputs created from it.

Yes.

• DAA is off by default. It is only enabled if you choose to opt in.

• DIA can be excluded from your service on request. Contact ai@microform.digital to discuss this. Without DIA, your documents will not be automatically tagged, categorised or clustered, so search and browsing will be less effective.

We have worked on AI and machine learning projects since 2018, in partnership with Sheffield Hallam University and UK Research and Innovation (UKRI). Through a Knowledge Transfer Partnership with Sheffield Hallam, we developed machine learning tools for historic archives, including image recognition for document analysis and natural language processing for intelligent search.

Read the case study: Helping Microform leverage the latest technology to drive business performance (Sheffield Hallam University)

Yes. Questions about our AI services, data handling or security are answered by our technical team, led by our Technical Director. Email ai@microform.digital or ask your account manager to arrange a call.