What the data for this category looks like
Intelligent due diligence report data for building construction engineering mainly comes from project bidding documents, construction design documents, cost lists, site visa forms, completion settlement reports, and similar materials. Data is updated as the project phase progresses, generated gradually after project initiation, and finalized and archived upon project completion. Most documents are multi-page PDFs, mixing structured tables, technical description text, and drawing screenshots. They include fields such as building area, project cost, material specifications, and construction duration, with common units including square meters, yuan, days, and millimeters.
Constraints on Document Parsing and Chunking
The multi-page mixed structure of building construction due diligence documents requires retaining cross-page table and text associations to avoid splitting the same bill of quantities entry. Nested structured tables contain multi-unit fields; standard parsing processes easily lose hierarchical relationships and unit information. PDFs with digital signatures may have encryption or signature layers that block basic OCR reading workflows. Documents are updated across project phases, with significant field differences between versions, requiring association with version identifiers. Long technical description paragraphs have vague boundaries, requiring identification of natural segmentation logic combined with industry-specific terminology.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the content density of long technical descriptions and structured tables in building construction documents, avoiding semantic breaks caused by excessive nested tables in a single chunk |
chunkOverlap | 100–150 characters | Retains associations between cross-segment bill of quantities entries, avoiding disruption of continuous pricing item descriptions during splitting |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to the parsing time required for multi-page large PDFs, preventing interruption of complex document parsing processes due to timeout |
enable_table_parse | Enabled | Accurately extracts hierarchical and field information from nested bill of quantities, avoiding table content being scattered into unordered text |
UPLOAD_FILE_MAX_SIZE | 500 MB | Accommodates single large completion settlement report documents, meeting document upload requirements for full-cycle due diligence of building construction projects |
enable_ocr | Triggered based on document type | Enables OCR for text extraction for PDFs with signatures or embedded images, disables OCR for plain text PDFs to improve parsing efficiency |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require case-by-case analysis, and it is recommended to test on your own samples before finalizing.
Three Common Misconfigurations
- Phenomenon: After uploading a PDF with a digital signature, the knowledge base parsing result is empty or only contains a small amount of text. Cause: The digital signature layer blocks the text reading path of basic OCR, and no targeted OCR triggering rules are configured.
- Phenomenon: A file read error is returned when calling the Doc2x tool, and the error report includes the "file read failed" field. Cause: The locally deployed pdf-marker service is not correctly connected to the FastGPT container network, preventing FastGPT from accessing the parsing service port.
- Phenomenon: After uploading a large cost list PDF, the nested levels of table entries in the parsing result are missing. Cause: The
enable_table_parseconfiguration is not enabled, and standard text parsing scatters nested tables into unordered text blocks.
How to Verify Correct Configuration
- Upload a signed building construction PDF, check whether the parsing result contains complete tables and text content, and confirm that the OCR configuration is correctly triggered based on document type.
- View the knowledge base parsing logs to confirm that no timeout errors are triggered, and adjust the timeout configuration to match the actual parsing time of the document.
- Check the segmented text blocks to confirm that the nested levels of the bill of quantities are fully retained, and verify that the chunking configuration adapts to the current document structure.
- Upload a single large completion report matching the project scale, confirm that the upload and parsing processes are not interrupted, and verify the rationality of the upload size configuration.
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.