AI
Workspace
AI · Company knowledge · Workflows
Solution map / AI WorkspaceDocumentsKnowledge baseWork tools
↓
MIRWINK / AIWorkspace↳ Team question
Answer with sources
↓
Approval → action↗
MIRWINK designs AI workspaces for documents, enquiries and internal tasks. We connect approved sources, define the assistant's role and link it to the tools the job requires. For each workflow, we establish what the system may do on its own and where a person makes the decision.
See how it works↓
AI · Company knowledge · WorkflowsMIRWINK ↗
When it makes sense
Close the gaps
in everyday work.
01 /The answer exists, but it is hard to find
Instructions, agreements and service information live across different documents. People open several sources and still need to check which version is current. We start by selecting the information your team can rely on.
02 /Repeated questions take up the working day
The team keeps explaining the same process or drafting similar replies. We identify those requests and build a workflow that finds the relevant information and prepares a draft for someone to review before sending.
03 /Context gets lost between a conversation and a task
A meeting or message thread leaves a long history to work through. We design a summary that separates decisions, open questions and next steps so someone can check it and use it in the next stage of work.
04 /A general AI chat does not know your rules
An answer may sound convincing without reflecting your services, constraints or procedures. We define instructions, approved sources and what to do when information is missing: ask for clarification or hand the question to a person.
05 /Automation needs clear permissions and ownership
Some documents are restricted, and actions in business tools need oversight. Before connecting tools, we define roles, access boundaries and the operations that require a person's approval.
Practical use cases
Your challenge.
Your team.
01An assistant for internal knowledge
A team member asks about a process, service or document. The assistant searches approved material and identifies the source when the chosen connection supports it. If information is insufficient, the workflow asks for clarification or routes the question to its owner.
↗02Enquiry classification and reply drafts
The assistant identifies the subject of an enquiry, extracts relevant details and prepares a reply using company information. A team member checks the draft, resolves anything uncertain and decides what to send to the customer.
↗03Document and conversation summaries
We turn supplied text into a summary of key points, decisions, open questions and possible next actions. The team agrees the format; ambiguous conclusions remain subject to human review.
↗04Preparing actions in business tools
An assistant can prepare the information for a task, record or another agreed action. We connect through the interfaces available in the target system. Where approval is required, a person first sees exactly what is about to change.
↗
The scope
What we
build together.
↗A defined workflow and quality criteria
We document who the assistant supports, which requests it handles and how its output is accepted. We also identify situations that must return to a person.
⊞Connected knowledge sources
We configure the agreed documents and available connections. We identify content owners, an update process and how outdated material should be handled.
↔Instructions and response formats
We define the assistant's purpose, output structure, tone and handling of uncertainty. Different tasks receive separate instructions where one workflow is not sufficient.
↳Roles and access rules
We agree who can use the workspace, which sources they may access and which actions are permitted. We check how those boundaries are supported by the chosen platform and connected systems.
⊕Tool connections and approval steps
We connect the workflow to agreed tools where the required APIs and permissions are available. We define approvals, error notifications and how exceptions are handled manually.
✓Checks and team handover
We check the workflow against an agreed set of routine and difficult requests. The handover explains how to use the assistant, update its sources and review answers that do not meet the criteria.
The path to launch
From a task
to a working flow.
- 01
Choose a specific task
We examine the current process: where requests arrive, where people find information and what the result should look like. We select the first workflow and agree how it will be checked.
→ - 02
Review data and permissions
We examine the sources, their accuracy and restrictions on their use. We identify content owners, connection options and the limits of the assistant's authority.
→ - 03
Build and check the workflow
We configure instructions, sources and tools. We compare answers with source material and check missing information, connection errors and the handover to a person.
→ - 04
Introduce it to the team and evaluate
We show the team how to use and review the assistant. We record observations and agreed measurements, then decide what needs to change before the workflow is expanded.
✓
Measuring the outcome
Progress you
can measure.
Before rollout, we agree the measures and baseline. Afterwards, we compare comparable periods and record the change. Values will follow the measurements.
01 / Reply preparation time—min
Compare median reply preparation and review time before and after implementation using comparable requests. Include time spent making corrections, and record the measurement period and sample alongside the result.
Measurement period —
02 / Drafts accepted without substantive changes—%
Measure the share of reviewed drafts accepted without changes to facts or meaning. Agree acceptance criteria in advance and disclose how many replies were reviewed.
Measurement period —
03 / Answers supported by approved sources—%
Check an agreed sample to establish whether material claims are supported by approved, current sources. Record refusals and clarification requests separately so they do not distort the result.
Measurement period —
Questions & answers
Before
we begin.
01What does AI Workspace mean in your approach?
+
It is a workspace where assistants carry out defined tasks using approved knowledge and tools. Our approach includes workflows based on Dust. We choose a suitable configuration after reviewing the process, information sources and permissions.
02Do all our documents need to be organised first?
+
The first workflow needs a limited set of current, dependable material. We identify that set, its owners and its update rules together. A large, unprepared knowledge base should not be connected without checking its content and permissions.
03Who will be able to see internal information?
+
We agree user and assistant permissions before connecting data. We check the boundaries supported by the platform and each source. Data handling requirements are discussed separately: a feature being available does not mean that every type of information is suitable for that service.
04Can the assistant make mistakes?
+
Yes. We therefore define approved sources, check answers against an agreed sample and plan for missing information. Important replies and actions require human review where agreed. Neither a knowledge base nor instructions can guarantee that errors will never occur.
05Can AI connect to our CRM and other tools?
+
This depends on the APIs, connections and permissions available in each system. We first define what the assistant needs to read and which actions it should prepare. We then agree approvals and error handling, assessing each connection individually.