AI applications
Search internal information, summarize material and help people work through incoming requests with sources they can check.
- Find relevant material
- Show the supporting sources
- Keep people in control of the answer
AI tools built around your data and systems.
Minutehouse designs and builds AI-enabled software, integrations and internal tools. That can mean a search tool grounded in approved documents or a connection between an order and a support system.
Illustrative example
Customer question
Delivery request
Can I change the delivery address?
The answer depends on the order status and the delivery policy.
Response workspaceReview required
A person decides what to send.
Information to check
Order statusFrom the order system
Delivery guidanceFrom the approved policy
The sources and permissions are agreed for each application.
Which tasks should software handle?
A support request may need an answer from several systems. A finance team may spend hours moving figures between files. We examine that work first, then design a tool or automation around it.
The right solution may use a language model, a conventional integration, or both. We agree on the data, permissions and review steps before building.
See how a project takes shapeWhat we can build
From a single workflow to the software around it.
Systems and integrations
Move information between applications, coordinate events and reduce the work of entering the same details twice.
- Agree which system owns each record
- Check updates as they arrive
- Route exceptions for review
Software and support
Build internal tools and web applications, then maintain the connections and processes they depend on.
- Build around the actual task
- Test common and exceptional cases
- Maintain software after release
Sources beside the draft.
When an application responds from company documents, a reader should be able to find the passage behind its answer. We design the interface and permissions around that check.
Read about reviewing AI outputTry an AI or integration task.
This example shows three possible uses of automation. Choose a task to see what a team would need to decide before it is put into use.
Illustrative workflow
Find the right answer for a support request
Identify the approved sources, retrieve relevant passages and let a person review the response before it goes to a customer.
Request receivedSources checkedResponse reviewed
Information usedApproved support material and the customer request.
Review pointA team member checks the response before it is sent.
From a task to a working application
Each project starts with the task, the information it needs and the checks required before release.
Define the task
Choose the request or decision the application should support, and identify who will use it.
Work with the sources
Agree which records are available, who can access them and how the application should use them.
Check and release
Test the expected cases and the exceptions, then decide where a person needs to review the result.
Check the rule and the current record
A delivery policy can explain when an address change is allowed. The order system supplies the current dispatch status. An AI-assisted response needs to show which source supports each part of the answer.
We distinguish a draft response from an instruction that changes a record. The application must show when a live check is unavailable and which action still needs a person’s decision.
Read about policy and live system checks