Services for institutions & newsrooms
Custom tools & chatbots grounded in your data
Make your documents and data usable in a purpose-built interface—with traceable sources, defined access and an honest answer when the evidence is missing.
Who & why
Your archive should support the work, not become another search task.
For newsrooms, research organisations and institutional teams whose knowledge is spread across files, datasets and internal repositories. A searchable collection, an extraction tool or a document-grounded assistant can serve different needs; a chatbot is not always the right interface.
We begin with the questions your team actually asks and inspect a permissioned sample. We then propose a tool around those tasks, including how users check the original source and what happens when documents conflict, are out of date or cannot support an answer.
Can AI-assisted material be published without an editor’s review?
Under the fictional document used in this example, material requires an editor’s approval before publication. No information about exceptions is available here; refer to the editorial lead.
Fictional reference: sample editorial policy, “Pre-publication review” section. Not a client document.
Retrieval and source references do not eliminate incorrect answers. A prototype demonstrates a bounded use case; production reliability, security review and ongoing operation need their own acceptance criteria.
Deliverables
Proposed deliverables, agreed before work begins.
Turn your organisation’s documents and data into custom tools and chatbots, with source references, access permissions and retrieval-quality testing.
A data-readiness and product brief
User tasks, source inventory, file quality, access rules and a recommendation for search, structured extraction, chat or another suitable interface.
A testable prototype
An agreed set of sources connected to a working interface, with source references and explicit handling of unsupported questions.
A retrieval and answer evaluation set
Representative questions, expected source material and review criteria for relevance, citation support, omissions and permission boundaries.
An agreed handover package
Documentation and, where contracted, code or deployment configuration, refresh procedures, operating requirements and responsibilities for maintenance.
How we work
A clear question. A reviewable handover.
Inspect the material
Review a sample for format, OCR needs, ownership, sensitivity and whether the required answers are present at all.
Design the access and retrieval path
Agree document preparation, metadata, permission checks and how the interface presents evidence or missing information.
Build and test with users
Implement the bounded prototype; test real tasks, misleading questions, conflicting material and attempts to access restricted content.
Decide what goes into operation
Review findings and agree whether to revise, stop or proceed, with explicit hosting, monitoring, update and support arrangements.
What you supply
What helps us scope a useful project?
- The intended users and examples of questions or tasks they need to complete.
- A small representative document or data sample you are authorised to share.
- Document ownership, sensitivity levels, access roles and any retention restrictions.
- Existing systems, hosting constraints and the person responsible for future content updates.
FAQs
Before we start a conversation.
Does the chatbot use the public web or our documents?
That is an explicit scope decision. A document-grounded tool can be limited to the approved collection. Public-web access is not assumed, and answers should distinguish source evidence from any other permitted information.
What if a document does not contain an answer?
We design and test an explicit no-answer response and a path back to source material or a human contact. A source reference alone is not proof: evaluation must check whether it actually supports the answer.
Can different users see different documents?
Role-based access can be included, but requires defined identities, permission rules and testing across retrieval, citations and exports. We do not assume a shared chatbot is appropriate for confidential collections.
Who owns and runs the finished tool?
Ownership, code access, hosting, provider accounts, usage costs and maintenance are agreed in the proposal. A prototype does not automatically include production hosting or an ongoing support service.
Let’s define a workable scope
What should your team be able to find or do?
Describe the task, file types and intended users. Start with a description of the collection—not confidential documents—and we can agree how to review a safe sample.