Our solution
Become an agent-powered knowledge organisation
Let our agents work with the information and expertise you already have.
NXT starts with your own material — documentation, manuals, product information, standards and expert knowledge. We structure it into knowledge areas and let NXT's specialised agents work with the information to create useful answers, training, tests, documentation and other content.
You own the knowledge. NXT's agents make it usable.
YOUR MATERIAL
Dokumentation
Manualer · instruktioner · standarder
Produktkunskap
Produktdata · specifikationer · ändringar
Expertis
Rutiner · erfarenhet · arbetssätt
Terminologi
Era facktermer · era begrepp · era språk
NXT AGENTS
NXT works with your material and knowledge context.
ASSIST
Asks · answers · guides
STUDIO
Explains · structures · teaches
LEXIKON
Translates · adapts · ensures consistent terminology
TESTCENTER
Tests · verifies · certifies
COMPLIANCE
Checks · documents · ensures
DOCS
Structures · publishes · makes accessible
RESULT
Knowledge you can use
Frågebara svar
Svar direkt ur era källor, med hänvisning till materialet.
Kurser
Utbildningar byggda på samma aktuella kunskap.
Tester & certifiering
Kunskapskontroll och intyg.
Compliance & dokumentation
Strukturerat stöd för krav, dokumentation och efterlevnad.
Tillgänglig information
Rätt information på rätt språk, för medarbetare, partners och kunder.
From documents to vectors
Knowledge is indexed by meaning, not only by words.
The foundation of NXT is a vector database built on your own material. It is what lets the agents find the right section even when the question is phrased in other words, or in another language, than the source.
- 01
Ingestion and normalisation
The source material — PDF, Word, web pages, product data — is read in and normalised. Heading levels, tables, lists and page references are preserved, because structure carries meaning.
- 02
Segmentation with context
The text is divided into defined sections that follow the document's own structure, not a fixed number of characters. Each section keeps its heading path, source, version, language and product affiliation as metadata.
- 03
Embeddings
A multilingual embedding model converts each section into a vector — a position in a high-dimensional space where proximity corresponds to similarity of meaning. The same model is used for both material and questions, so they end up in the same space.
- 04
Storage in a vector database
The vectors are stored together with the original text and metadata in a vector database within the EU. A vector index keeps the proximity search fast even across large knowledge areas.
Multilingualism
Any language, one meaning — without translation in between.
The multilingual embedding model places “Hur byter jag filter?”, “How do I replace the filter?” and “Wie wechsle ich den Filter?” close to one another in vector space. The question therefore finds the source section without being translated first. The capability sits in the model, not in the database, and the answer is then formulated in the questioner's own language. Quality varies with technical terminology and compound words — which is why we work on the terminology in LEXIKON and measure retrieval accuracy against real questions.
Search and answers
Semantic search, exact matching and a source requirement — together.
Semantic search
The question is converted into a vector and compared with the sections using cosine similarity. This means “the machine stops at start-up” can find the section on “emergency stop triggered during start-up”, even though the words differ.
Exact matching where it is needed
Part numbers, error codes and standard designations have no “meaning” in vector space. They are matched exactly using text search, and the results are weighted together with the semantic search.
Filtering on metadata
The search is limited to the right knowledge area, product, version and permission before any answer is formulated. A user never sees content they are not entitled to.
Answers with a source — or no answer
The language model receives only the retrieved sections as its basis and states the source. If the material is not enough, the system says so, instead of filling the gap with guesses.
The agents
Specialised agents work against the same vector database.
Each agent has a defined task. One structures the material, another secures the terminology, a third creates content. All of them draw their basis from the same index, which means that answers, courses and tests build on the same version of the knowledge.
DOCS
· StructuresOrganises the material into knowledge areas, headings and versions, so that the vector database has a clean and traceable basis to work against.
LEXIKON
· Secures the terminologyKeeps technical terms, synonyms and language variants in order, and ensures that the same concepts are used consistently in everything that is created.
STUDIO
· Creates contentRetrieves relevant sections from the vector database and builds courses, explanations and exercises on them — with traceability back to the source.
TESTCENTER
· Verifies knowledgeProduces questions and tests from the same basis, so that what is assessed matches what is taught.
COMPLIANCE
· Reviews against requirementsCompares the content with applicable requirements and standards and shows where supporting material is missing.
ASSIST
· AnswersAnswers questions in real time with a reference to the source, and says when the answer is not in the material.
Security and control
Data at rest within the EU
Material, vectors and metadata are stored within the EU.
Access-controlled search
Access is filtered already in the search, not only in the answer.
Traceability
Every answer and every generated section can be traced back to the source.
A person approves
Content created by the agents is reviewed by a subject-matter expert before it is published.
One source. Multiple agents. Many uses.
The same knowledge does not need to be recreated every time it is used.
A product manual can become a source a service technician can ask questions of, a course for a dealer, training for a customer, a test for an operator and documentation at the machine.
That is what we mean by Knowledge, activated.