ProfITsoft embeds artificial intelligence not only into the process of building KSASK, but into the product itself — where an insurance-company employee works with it every day. The goal is simple: take routine answer-hunting, «configure-by-example» setup and manual checks off the specialist’s plate so they focus on decisions, not mechanical work. The direction covers answering user questions, an assistant inside the system interface, and intelligent features within the insurance business processes themselves.
AI support assistant
The AI support assistant helps KSASK users. Instead of searching the help or writing to support and waiting, a user asks a question in plain language — and gets an answer in seconds, with a link to the source in the documentation. If a question is complex or outside the knowledge base, the assistant hands it to a human — a support moderator who can clarify or extend the answer. So the client always gets a correct answer, and the support team is relieved of repetitive routine requests.
How it works
At the assistant’s core is our own knowledge-base service using RAG (retrieval-augmented generation). It does not invent an answer «out of thin air» but relies on the company’s uploaded documents: upload documents → semantic indexing → retrieve relevant fragments by meaning → re-rank by relevance → generate the answer with a language model, citing sources.
- Semantic search, not keyword search. The assistant finds what is relevant by meaning, even if the wording does not match the instruction text literally.
- Answers with source links. Every answer includes the sources it was assembled from — the user can verify and dig deeper.
- Works in many languages. The text-understanding model is multilingual — Ukrainian is supported natively.
- Resilient to provider outages. Several language models are supported with automatic switching — if one provider is unavailable, another serves the request.
- An up-to-date knowledge base. Documents are synced from sources and re-indexed on changes — the assistant answers from the current revision.
The AI assistant inside KSASK
The AI assistant is embedded directly into the KSASK interface, next to the usual screens. AI components are added alongside the back-office, without changes to the system core, and data does not leave the client’s perimeter.
- A chat widget in the familiar interface. A floating button and a side panel are embedded into existing pages — no separate application.
- Actions on the user’s behalf. When the assistant performs an action, it calls existing system functions with the user’s own permissions — no more than the employee has. Each action has a mode: read-only (runs automatically) or write (with the user’s confirmation).
- A knowledge base under the client’s control. The insurer’s administrator uploads instructions, internal procedures, regulations (NBU, tariff changes), typical «question → answer» pairs and the assistant’s behaviour rules.
- Roles and control. Separate access rights: using the assistant, reviewing/moderating answers, administering tools and rules.
Intelligence in business processes
A separate direction is intelligent features directly within insurance business processes, where AI does not just answer questions but helps make decisions. An example is AI risk assessment during claims settlement: hints to the specialist while reviewing a case, signals about atypical indicators that need closer checking, a summary of the case materials, decision support. The decision always remains with the human; AI prepares the groundwork.
Benefits for insurers
- Faster service and less routine — typical questions and setup are handled in seconds, the specialist focuses on complex cases.
- Less load on support — the assistant takes on repetitive requests, a human steps in only where truly needed.
- Answers on your data — the assistant relies on your company’s own documents and rules, with source links.
- Data under your control — the architecture is designed to work within the client’s perimeter; only a de-identified request goes out to the model.
- ProfITsoft’s single AI competence — the same experience that speeds up KSASK development also powers AI inside the product; the direction grows together with the Functionality Exchange Fund.
Security and data control
- Working within the client perimeter. AI components (assistant, knowledge base, storage) are deployed alongside the back-office; dialogue history, documents and settings do not leave the company’s infrastructure.
- De-identification before sending to the model. The only external traffic is requests to the cloud language model; personal data is masked before sending.
- Your data is not used for training. Model providers are configured so that data is not stored for training.
- No permissions beyond the user’s. The assistant acts with the employee’s own permissions and gains no additional rights.
- A fully local model option. If policy forbids any external traffic with data, the model can be deployed locally within the client’s perimeter.
Technology
- RAG architecture — answers are built on the company’s documents, with citations, rather than «made up».
- Semantic indexing and re-ranking — search by meaning and selection of the most relevant fragments before forming an answer.
- Several language models with failover — leading cloud LLMs with automatic switching on outages; local deployment when required.
- Deployment alongside the system — the AI stack is added to the client’s existing infrastructure with no changes to the KSASK core; integration through existing authorisation mechanisms and APIs.
Connection with the «AI in development» direction
AI inside the product is a continuation of our engineering AI competence. In parallel, ProfITsoft applies AI systematically in developing KSASK: from forming requirements and writing code to automatic review, documentation and testing. It is this accumulated experience that lets us confidently build AI features into the product itself.
Want to learn more?
We will explain in more detail how the AI assistant helps insurance-company users and how intelligent features evolve within KSASK.