Custom software · Produkt + Markt
Qualytis
- Starting point
- Qualitative interviews need to be coded systematically.
- Application
- Custom software that supports the coding of qualitative interviews.
Service · Custom development Custom AI applications for your team.
We build applications that help your team analyze data, unlock the knowledge in documents or automate recurring work steps.
Not sure yet whether deepsight cloud, slide factory or custom development is the right fit? Then validate your use case with us first.
Common hurdles
Rules in handbooks, details in PDFs, workflows in spreadsheets.
And in between, your team searching, retyping and cross-checking. For an AI application to help, it has to fit your data, systems and requirements.
The challenge: The answer is in there somewhere
Policies, contracts, service descriptions: anyone looking for a rule opens several documents – or asks the colleague who knows.
With deepsight: Answers with a source
A knowledge chatbot searches the documents you approve and names the passage it relies on. Your team checks the source instead of searching.
Go to knowledge chatbotsThe challenge: Retyping, day after day
Invoices, applications, forms: the same details are keyed into your business system by hand. Typos often only surface later.
With deepsight: Extract fields, flag what’s uncertain
Data extraction captures the details you need. Values that are hard to read or break an agreed rule go to your team for review before they reach the target system.
Go to data extractionThe challenge: A workflow no standard tool covers
Your own method, your own checks, your own report format: whatever the software can’t do ends up in spreadsheets on the side. Export after export.
With deepsight: Software built around your method
We build the application around your workflow: with your terminology, your checks and the results your team works with next.
See project examplesThe challenge: The pilot that never goes live
The prototype is convincing. Then IT asks: where is the data processed, who runs the system, how does it connect to what’s already there?
With deepsight: Operation planned from day one
We settle data flows, operation and integration with your IT early – not after the prototype. The prototype, the build and, if you wish, operation stay with the same team.
See the project phases
Where we help
Three kinds of tasks
Analyze dataYour own analysis and reporting logic, for example for coding qualitative interviews.
Unlock your documentsKnowledge chatbots for your documents and archives.
Automate work stepsData extraction and integration with your systems.
Local language models: We set up, maintain and operate local LLMs – in your own infrastructure if you prefer.
Need a custom-trained model?
Custom-trained modelsAs part of a custom project, we assess whether this approach suits your use case.
Projects
More about our research projects Starting point, application, status
Working together
Phases and what you get in each
- Clarify
Use case, data and stakeholders. You receive a requirements description with an effort estimate.
- Prototype
A first version using a sample of your data that your team can test.
- Build
The application ready for use, connected to your systems.
- Operation
Hosting, support and further development as agreed.
Which models, databases and interfaces we use depends on your project and your IT. We define the infrastructure for each project – including local language models in your environment.
We work with widely used technologies and documented code. Usage rights and handover are set out in the project contract.
Questions
What an application can answer or handle for your team
- Which rule applies to this case under the current policy – and where does it say so?
- What does our study or project archive already hold on a topic?
- Which details from this application form go into which field of our business system?
- How does our team code qualitative interviews consistently against our own guide?
- How do we run an entire survey as an accessible conversation rather than a questionnaire?
- Is a text easy to understand, and where are the language barriers?
Good to know: In deepsight cloud, surveys with an AI follow-up can ask a follow-up question when an open-ended answer is brief, and the assistant answers questions about your analysis – neither needs a development project.
Data and sources
What projects start with
Usually the data already exists – spread across drives, mailboxes and business systems. A typical sample is enough for the prototype.
DocumentsPolicies, handbooks, contracts or study reports, for example as PDF or Word files.
Recurring incoming documentsInvoices, applications, forms or delivery notes, including scans.
Interviews and surveysTranscripts of qualitative interviews, open-ended responses or conversation logs, with your interview guide or code frame.
Data from your systemsTickets, cases or records from the systems your team works with.
Your expertiseTerminology, validation rules and the cases where even your team has to look twice.
Good to know: You can remove or mask personal information in the sample before handing it over.
Your own data shows whether an AI application holds up.
Sample data says little about your scans, your terminology and your edge cases.
Process
How we test the prototype
The “Prototype” phase step by step.
- Collect test cases
Your specialist team puts together typical and difficult cases: real questions with the correct answer, documents with the correct values, or interviews your team has already coded.
- Compare results
We hold the prototype’s results up against these cases: where does it get it right, where not – and is that down to the documents, the document quality or the method?
Example: measuring a sample - Refine
Add to the documents, adjust validation rules and define which values go to manual review. Then compare again.
Example: review view - Try it in daily work
A few users work with the prototype on real tasks. Their feedback goes into the requirements for the build.
- Decide
On this basis, you decide whether the application is built and to what extent.
Results
Deliverables we agree for your project
Prototype evaluationIf prototype testing is agreed, we document how the prototype performed on your test cases.
DocumentationWe agree the scope and depth of the documentation covering architecture, data flows, access and operation for your project.
OnboardingWe agree with you whether onboarding is included and who it is for, such as users or everyone who will maintain documents and rules later on.
You decide what the application may do. Not the language model.
Which documents it uses, which steps it takes and when a person confirms.
Example: actions with approvalMethodology
What we pay attention to in AI applications
Combine language models with fixed rules
Language models read free text and varying layouts. Fixed rules check what is clear-cut – such as required fields, formats, totals or the permitted codes of a code frame.
Keep reference material current
An application is only as current as its reference material: documents, validation rules or code frame. Before launch, we agree who adds new versions and who retires old ones.
Say so, don’t guess
If an answer isn’t in the documents, a value is illegible or an interview passage fits no code, the application should show that openly. How reliably that works is part of the test cases.
Start narrow, then expand
First one clearly defined case, such as one document type, one topic area or one study. More follow once the first one works in daily use.
FAQ
Questions about custom projects
There is no flat rate: effort and duration depend on data, integrations and operation. Both can only be estimated reliably once the task and the data are clear – which is why the estimate comes before any development.
For analyzing open-ended responses, surveys or feedback, deepsight cloud is the direct route; for Excel tables that need to become PowerPoint slides, it’s slide factory. Custom development makes sense when your task, data or systems need something neither of them covers.
Above all, suitable hardware, usually servers with graphics processors (GPUs). What’s enough depends on the model, the data volume and the number of simultaneous users. Our experience ranges from GPU servers in the data center to compact AI workstations. The prototype shows whether a local model is sufficient for your task.
Yes, if your systems offer an interface or an export – for example a CRM, ticketing system, database or business application. We agree on the route and the application’s permissions with your IT. Integrations with deepsight cloud that go beyond upload and connectors are also built as a custom project.
Which data goes to which services and who has access is clarified at the start of the project and built into the architecture. If we process personal data on your behalf, we sign a data processing agreement (DPA) under Art. 28 GDPR. The information under Security refers to deepsight cloud and slide factory.
Further reading
Related to custom projects
EnterpriseClarify scope, integration and operation with business units, IT and procurement in one joint briefing.Go to Enterprise ResearchTOP.KI and the survey chatbot in detail: goal, our contribution, funding and status.Go to research
Discuss your project
Tell us about your task. We will get back to you with an initial assessment.