Service (003)
Data & Knowledge Systems
Make what your company knows findable.
The answer is in a contract from 2019, a shift log, or the head of someone who retires in March. We build knowledge systems that find it, quote the source and respect who is allowed to see what.
- Knowledge search (RAG)
- Data readiness & pipelines
- Evaluation & testing
- Analytics & forecasting
- Private & on-premise deployment
When it fits
You probably need this if any of these sound familiar.
People ask colleagues instead of searching
Because search returns four hundred files and none of them is the right version.
Expertise is walking out the door
Your most experienced people are near retirement and their knowledge is not written down in any usable form.
You cannot send data outside
Legal, medical or client confidentiality rules out public AI tools.
Reports take days
Every management question becomes a manual export and a spreadsheet.
What you get
Deliverables(6)
- (001)
Knowledge search with citations
Answers in plain language, each one linked to the paragraph it came from.
- (002)
Permission-aware access
The system only retrieves what the person asking is already allowed to open.
- (003)
Data pipelines
Documents and records cleaned, structured and kept in sync with the systems of record.
- (004)
Blind-reviewed evaluation
A question set written and scored by your own experts, so accuracy is a measured figure, not a feeling.
- (005)
Private deployment
Hosted in an EU region, in your own cloud tenancy or on your own servers, depending on your rules.
- (006)
Usage analytics
What people ask, what the system cannot answer yet, and where the documentation has holes.
How it runs
Timeline
- Weeks 1–2
Inventory
What exists, where it lives, who owns it, who may see it and how much of it is junk.
- Weeks 3–6
Index
Parsing, cleaning, chunking and permissions for the first collection, with an expert-written test set.
- Weeks 7–10
Tune
We run the test set after every change until the experts sign off on accuracy.
- Weeks 11–16
Roll out
Team by team, with training, feedback buttons and a weekly review of failed questions.
- Typical length
- 3 to 5 months
- Team
- Data engineer, ML engineer, delivery lead
- Fee
- Fixed per phase
- You get
- A system that runs in your environment
(2)
Related work
Engagements where Data & Knowledge was part of the scope.
Good to know
Questions
Do we need to clean up our data first?
No. Finding out which data is good enough is the first phase of the work. Most clients are in better shape than they fear in two areas and worse in one.
Does our data train anyone’s model?
No. We deploy under terms that exclude training on your data, and for sensitive work we use models that run entirely inside your environment.
How accurate is it?
We agree a target with your experts before building and measure against it. On recent projects, between 88 and 94 percent of answers were judged correct with the right source in blind review.
What about documents in Hungarian, German or Czech?
Normal for us. Most of our collections are mixed-language and the evaluation sets are written in the languages your people actually use.
Other services
Have a project in mind?
Get in touch
Tell us what you are trying to fix, in a few lines. We reply within one working day with questions or an outline of how we would approach it.
Give us a call
Head office
Delivery studio