This piece is by Christos Avgerinopoulos, Independent developer of Asfalentra
An insurance renewal can look attractive because the annual premium has fallen, while other wording deserves closer attention. A deductible might have increased, a flood limit might have decreased, or a condition might have been added. For the professional reviewing the documents, the task is to establish what changed, locate the relevant passages and assess their significance in context. Asfalentra is a working Greek-language software pilot developed around that specific task: comparing a commercial-property policy with its renewal.
The project is being developed independently in Greece by Christos Avgerinopoulos. Its initial audience is Greek-speaking insurance intermediaries and brokerage teams handling business-property documents. It is currently seeking professional evaluation. It has no customer case studies or independently established operational results to announce. The immediate objective is to test whether the workflow is useful in practice, identify missed or unhelpful findings, and measure the time required for a complete review, including the professional’s own checks and corrections.
Users bring two documents into a local workspace: the existing policy and the proposed renewal. The application accepts text-based PDFs, UTF-8 text files or pasted text. It extracts the text, retains page references for PDF passages and presents textual additions, removals and replacements. A full extracted-text comparison operates alongside eight keyword-based thematic views: annual premium, building and contents, deductibles, flood, earthquake, business interruption, exclusions and special conditions. These views organise candidate passages rather than certify the meaning of the insurance wording.
A synthetic demonstration illustrates the intended use. The annual premium falls from EUR1,240 to EUR1,110, while the general deductible rises from EUR500 to EUR1,000. The stated flood limit falls from EUR100,000 to EUR50,000, with its deductible increasing from EUR1,000 to EUR2,500. The maximum business-interruption period changes from six months to three. An additional exclusion and a drainage-inspection condition also appear. These are invented demonstration figures, not a customer’s policy, an insurance quotation or evidence of a particular insurer’s practices.
The application presents those passages for a professional to examine against the originals. It does not decide whether the renewal is suitable, whether two clauses are legally equivalent, or whether a future claim will be covered. It does not recommend an insurer or calculate a payout. An unchanged passage does not prove unchanged overall protection, and a missing keyword match does not establish that cover has disappeared. Reordered wording can also appear as a removal and an addition, requiring the reviewer to check the relationship.
Optional AI assistance runs locally using an existing Qwen3 4B Instruct model. The model has not been fine-tuned for this project. It receives a bounded selection of candidate passages and can select up to three pairs for review. The software then inserts the original source text and checks the references, rejecting certain invalid selections. This is a limited passage-selection aid, not a complete AI assessment of both policies. The broader textual comparison and the professional’s examination remain separate parts of the workflow.
Reviewers can add notes, mark items as examined, save a case file and export a text or printable HTML report. The HTML can be printed to PDF through the browser. Saved case files contain readable document text and notes, so users must choose appropriate storage and access arrangements. Work must be explicitly saved before closing or refreshing the page. After installation, document processing takes place on the computer without sending policy contents to an external model service. Local processing is not, by itself, security or data-protection certification.
Evaluation would begin by confirming a participating office’s equipment, document formats and actual renewal-review process. Any use of real documents would need appropriate authorisation and handling arrangements. The proposed comparison is against an experienced professional’s review, recording missed changes, incorrect indications and the time spent checking and correcting the output. That evidence would determine whether the tool merits further development or a paid pilot, rather than treating technical test results as proof of business value.
The tested AI installation is for Apple Silicon Macs and Greek PDFs containing selectable text. Scanned-document recognition is not included; neither Windows AI operation nor an English document workflow has been validated. There are no insurer-platform integrations, team synchronisation or customer-account services. Technical checks support the pilot’s operation, but insurance accuracy and time savings still require professional evaluation. An initial demonstration uses synthetic documents through screen sharing. There is currently no public interactive demo; the linked form is specifically for requesting a demonstration. Any subsequent commercial pilot would be scoped and agreed separately.
Be the first to comment