AI that does a defined job inside your product
We add generative AI to systems people already work in, build search and answers over a company's own content, and advise where AI is worth using. Each solution has a narrow task, a person who approves the result and a known cost.
What the service covers
-
Generative AI in existing products
Helpers that draft, shorten, rewrite or describe content at the place where a person works on it, such as a button in an editor.
-
Search and answers over your content (RAG)
Your documents or articles are indexed by meaning. A question finds the relevant passages first, and the model answers from them.
-
AI consulting
We look at a process, estimate what a model would cost and save per item, and recommend the solution that suits the job best, even when that is one without AI.
-
Integration and operation
Connections to model providers, limits, logs and switches. The solution is secure, costs are predictable and stay within set limits, and usage is shown in clear reports.
Work
-
Ir.lv
Headline and intro variants, SEO texts, related articles by meaning and article audio for a weekly magazine.
Read the case study: Ir.lv -
Lasi.lv
AI helpers in the editorial CMS of a national publisher, with usage shown in the logs.
Read the case study: Lasi.lv -
E-klase
An education news portal for teachers, parents and students, with an editorial CMS and site search.
Read the case study: E-klase
Start with one task
An AI project needs a task that is small and can be checked, for example "write a search description of up to 155 characters from this intro".
We begin with one such task, test it on real examples from your content and agree what a good result looks like before anything is built into a product.
How search over your own content works
A language model knows nothing about your documents. Retrieval-augmented generation, or RAG, gives it the right passages at the moment of the question.
Each text is turned into a vector that describes its meaning and stored in a database. A question is turned into a vector the same way, the closest texts are found, and the model writes its answer from them. The same search, without the last step, finds related articles or similar products.
- Your content
- Vectors
- Vector database
- Closest passages
- Answer with sources
How we keep quality
A model writes fluent text whether or not it is right, so these controls are part of the design from the start.
-
Only the facts it is given
The model works from the text supplied to it and is told not to add names, numbers or claims of its own.
-
Reuse before rewriting
Where a human text is already good, the tool keeps it. A model is asked to write only where something is missing or weak.
-
A reason with every result
The tool explains its decision in a sentence, which makes a wrong result easy to spot.
-
A person approves
Results are suggestions inside the tool people already use. They can edit or discard them.
-
Tested on your examples
Before release we run the tool on a set of real cases of different kinds and read the results with your team.
How we keep cost under control
The cost of a language model depends on how much text goes in and comes out. We send the shortest context that gives a good result, generate again only when the source changes and cache what has been made.
Every call is logged with its duration and cost, and limits are set for the whole system and for each model. You see what each feature costs per month before it is switched on for everyone.
Not tied to one provider
Models and prices change every few months. We put one service layer between your system and the model providers. A provider can be changed in the settings, a backup takes over if the main one is down, and your system keeps working without the AI feature if none is available.
We agree in advance which data may be sent to a model. Only the text a task needs leaves your system.
Related pages
The AI tools we have built for newsrooms are described in detail on the expertise page.
Wondering where AI would pay off in your work?
Describe one task your team repeats every day, and we will tell you what we would try.