What the data for this category looks like
Data sources for general comprehensive intelligent due diligence reports include publicly disclosed enterprise documents, internal due diligence working papers, third-party credit reports, and project-related attachments. Data updates follow the progress of due diligence projects, with no fixed cycle. A single report contains multiple types of document structures: structured financial detail tables and compliance item lists, semi-structured due diligence description paragraphs, and unstructured scanned qualification documents. Fields include due diligence numbers, project entity names, compliance risk items, and quantitative business data, each with a clear unit. Some reports also link multiple attached documents.
Constraints imposed by these characteristics on the document parsing and chunking process
Mixed document structures require the parsing process to support structured table extraction, semi-structured paragraph splitting, and unstructured image text recognition simultaneously, to avoid missing key information. No fixed update cycle and large single report volume require the parsing process to support batch processing of large files, to avoid timeout during single parsing. The feature of field-unit binding requires retaining contextual association during chunking, and not splitting quantitative data and its corresponding descriptions into different chunks, as this will affect subsequent retrieval accuracy. Scenarios with multiple attached documents require the parsing process to automatically associate the main report and attached documents, and generate unified chunked content.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
enable_advanced_pdf_parse | Enabled | General comprehensive intelligent due diligence reports contain a large number of scanned qualification documents and attachments. Enhanced parsing can extract text embedded in images, avoiding loss of key information |
chunk_size | 800–1200 characters | The report contains both long paragraphs of compliance descriptions and short table items. This range balances contextual coherence and retrieval accuracy |
chunk_overlap | 100–150 characters | Retains contextual association across chunks, avoiding splitting quantitative fields with units and their corresponding descriptions into different chunks |
enable_table_vector | Enabled | Financial tables and compliance lists in due diligence reports are core retrieval content. The table structure must be converted into a retrievable vector format |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | A single report may contain multiple attachments. This duration covers the time requirements of most batch parsing scenarios |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapts to the upload and parsing requirements of single large due diligence reports and attached documents |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material format, data volume, and business rules. Specific issues require individual analysis. It is recommended to test on your own samples before finalizing the settings.
Three Common Misconfigurations
- When calling document parsing via the API interface, the returned parsing result does not include image embedded text. The cause is that the
enable_advanced_pdf_parseparameter is not correctly configured totruein the request parameters. - When parsing table data from due diligence reports, retrieval results only return scattered paragraph text. The cause is that the
enable_table_vectorconfiguration is not enabled, and the table structure is not converted into a retrievable vector unit. - When parsing a due diligence package with multiple attachments, a
504 Gateway Timeoutstatus code appears. The cause is that thePARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted based on the total file size, and the default duration is insufficient to complete the batch parsing process.
How to Confirm Proper Configuration
- Enter the settings page of the corresponding knowledge base, and check whether the
enable_advanced_pdf_parseswitch is in the enabled state. - Upload a test due diligence report containing scanned PDF attachments, and check whether there are related records of image text extraction in the parsing log.
- Upload a test document containing financial tables, perform a retrieval operation, and verify whether specific field content in the table can be retrieved.
- View the post-parsing chunk list, and confirm that no chunk splits quantitative fields with units and their corresponding description text.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.