Comparison of insurance documents with proof
Klemens.AI is a platform for automatic analysis and creation of insurance documents.
Klemens.AI compares dozens of documents at once: policy wordings, product information documents (IPIDs), offers and clauses, even in several languages at the same time. You get the results in the format you need. For each value, it provides the file name, page number, and verbatim quote. Where data is missing from the document, it replaces the closest matching fragment with a "—."
Klemens.AI will verify the answer before displaying it and will not show an answer that it cannot prove with a quote from your document.
This is not a document chat.
This is a tool for tasks where the result must be defended.
Who is Klemens.AI for?
For people responsible for a product portfolio across several markets, a product review under POG and IDD, post-merger harmonisation, or a comparison for group headquarters against a deadline. See details for insurance companies.
The result in the form you need
You decide how the system presents the analysis. A comparison table is the most common format, but not the only one. The same summary can take different forms, depending on who you show it to and why.
-
Comparison table: each product or offer times each agreed field.
-
List of differences between documents or between versions of the same document.
-
A list of gaps, omissions and inconsistencies, i.e. what is not in the document.
-
A set of questions for negotiation, derived from comparison.
-
A descriptive summary for management or the client, written in plain language.
-
Export to DOCX or PDF.
Regardless of the form, each value has an address in the source document, and missing data remains missing.
From analysis to the finished document
Analysis is rarely the end of the work . Usually, there's still some writing to do: a response to the client, a memo to management, a summary of the findings. Klemens.AI completes this second step in the same place, basing its content on the results of the previous analysis rather than on the model's general knowledge.
The document is created from material that has already been verified against sources, so its claims can be linked to a specific record. You finalize the finished text on the platform, review previous versions, and export it to DOCX or PDF.
You sign it, so before sending it, you read it just like any other document. We don't design a process where a letter goes out to a client without human approval.
LLM proposes, code verifies
Large language models are good at finding content, but they can sometimes make things up. Therefore, in Klemens.AI, the model only suggests an answer, and a separate layer of code checks it against the document before you see the result. Four rules always apply.
Proof for every value
Specific file name, page number, and verbatim quote. No paraphrasing, no summarizing, no vague "based on the document."
Honest "—"
Missing data is reported as missing. This is the most common source of incorrect decisions in comparisons, because a semantic search engine will always return something, even when the desired entry simply doesn't exist in the document.
Determinism
The same question asked again on the same set of documents yields the same substantive answer. The result can be repeated and shared with another person.
Distinguishing between error and missing data
The system separates the situation of "I didn't find it in the document" from "the analysis failed." For the person responsible for the result, these are two completely different pieces of information.
Security
Klemens.AI is auditable,. Your IT department can get administrative access to the environment and check everything themselves before you add your first document.
Limiting hallucinations
Klemens.AI mitigates the risk of hallucinations by basing answers on controlled sources and verifying each citation against the document. When data is missing, it reports it as missing rather than providing it.
We're not claiming that hallucinations have been completely eliminated. We're saying otherwise: any value can be verified in just a few seconds, because we know where to look for it in the document .
Security of your data
Each customer receives their own dedicated space in Google Cloud: their own database, their own file storage, their own indexes, and their own application instance. Web search is disabled by default. This isn't a shared database with filters, but a separate infrastructure.
Your data remains in the European Union . The application runs in Frankfurt, files and background processes in Warsaw, and LLM calls are made through a European endpoint.
Each response leaves a permanent record and is auditable : what the scope decision was, which documents were read, what quotes were made, what analysis coverage was, and what was included in the model context and what was removed. A separate event log is maintained: who changed what, when, and with what before and after the change. This is a completely different situation than a screenshot from a chat.
What doesn't Klemens.AI do?
-
It does not replace the person responsible for the result . The result is always verified by the specialist who signs it.
-
He does not issue a broker recommendation . He prepares material on which the broker bases his own recommendation. This is an activity covered by his professional responsibility.
-
It does not interpret the regulations and does not replace a legal opinion.
-
It doesn't send anything for you . You read and approve the document prepared on the platform before it goes to the client.
-
It does not promise maintenance-free operation without human supervision .