All cases
Studio case // LISVENT // 1C catalog

LISVENT: matching customer requests to a 1C catalog

Customers use familiar names, while the same products appear under different wording in 1C. For LISVENT we built a system that parses the request, retrieves catalog candidates and flags uncertain lines for a manager. Model numbers and dimensions are checked separately from text similarity.

Catalog
More than 30,000 product records
1C exchange
Catalog export and final Excel file
Decision
Search cascade and manager confirmation

What we built

A manager uploads Excel or pastes text, reviews candidates from the catalog and confirms the result. The 1C catalog enters through a file export; the result is returned in Excel. This workflow does not create documents directly inside the 1C database.

The operational workflow

  1. 01Select the customer and upload Excel or paste request lines.
  2. 02Review the parsed input and confirm processing.
  3. 03Check dictionaries and approved mappings before fuzzy and semantic retrieval.
  4. 04Give the model a limited candidate set; validate numbers and model identifiers in code.
  5. 05Confirm or replace suggestions and export the final Excel file.
Workflow diagramGERON LABS

LISVENT: file-based exchange

  1. Export the 1C catalogue

    An administrator uploads the catalogue and refreshes its search index.

  2. Excel or text request

    The parser identifies descriptions, quantities and units.

  3. Match and review

    The system suggests candidates; a manager reviews ambiguous rows.

  4. Final Excel file

    Confirmed items are exported for the next stage of work.

This project does not write documents directly to 1C and does not use OData.

The actual LISVENT integration route, shown as a workflow diagram rather than a product screenshot.

Why semantic similarity is insufficient

Similar ventilation products may have different dimensions, voltage or model numbers. Retrieval finds candidates but does not establish interchangeability.

We added explicit model-number retrieval. Code verifies model tokens and numbers after candidate selection. A mismatch sends the line to human review.

What needed correction

The Excel parser sometimes selected a quantity column as the product name. Header validation now requires a name and another meaningful column; numeric columns cannot silently become item descriptions.

Another fallback could restore the original AI suggestion after the manager rejected it. The export now uses an explicitly confirmed record. Rejected suggestions do not reappear through fallback logic.

Building reusable mappings

A confirmed customer phrase and catalog record enter a review queue. An administrator can approve, replace or reject the pair.

Approved mappings help with later requests. A single accidental edit does not automatically become a permanent rule.

What changed for the manager

The manager sees prepared candidates and statuses rather than starting every line from scratch. They resolve ambiguous items, check units and quantities and accept the final request.

File exchange allowed the workflow to use the existing catalog without changing the 1C configuration. OData or custom HTTP integration can be designed separately for a defined business operation.

Sources and links

SYS: COOKIE

We use technical browser storage so the site works correctly. Privacy Policy