What the data for this category looks like
The business data for building engineering general contracting mainly comes from project construction logs, general contracting contracts, progress approval forms, material inspection slips, and completion archiving materials. Data is updated according to project construction phases. Progress approval materials are submitted weekly. Progress payment declarations are made monthly. Single project archiving documents often contain multi-chapter text descriptions, Excel bills of quantities with multiple embedded sheets, and PDF-format on-site image attachments. Document fields include project number, section division, construction location, measured project quantity, material model, inspection date, and other items. Units use engineering-specific measurement standards such as cubic meters, square meters, tons, and labor days.
What constraints do these characteristics impose on the document parsing and chunking workflow
Documents for building engineering general contracting have embedded multi-format attachments and multi-sheet tables. General parsing modules may miss content from non-default Excel worksheets. They may also mistakenly trigger OCR for PDF-embedded images, generating redundant text and damaging the original document structure. Single documents have long length. Parsing time increases with document size. Without adjusting the timeout threshold, parsing failures are likely to occur. Engineering-specific fields and measurement units must retain their original format. General parsing logic may incorrectly convert non-standard engineering terms. This affects the semantic integrity of subsequent chunks.
How to configure the settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 1000–1500 characters | Ensures semantic integrity for long paragraphs and embedded tables in building engineering documents, avoids splitting complete statements related to engineering measurements |
chunkSplitter | Split by paragraph + heading | Matches the multi-chapter structure of documents, retains contextual association for fields such as construction location and project quantity |
customPdfParse | Disable OCR enhancement mode | Prevents accidental conversion of PDF-embedded on-site images into redundant text, preserves the original image attachment structure |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Adapts to parsing time for large single engineering documents, prevents parsing failures caused by timeouts |
UPLOAD_FILE_MAX_SIZE | 500 MB | Adapts to general contracting project archiving documents with multiple attachments, prevents upload truncation |
chunkSettingMode | Manual configuration | Adjusts chunking rules for engineering-specific fields, avoids generic mode incorrectly splitting content related to measurement units |
Three common configuration mistakes
- Phenomenon: When parsing PDF documents, embedded on-site images are converted to OCR text, and the original images are not retained. Cause: The OCR enhancement mode of
customPdfParseis not disabled. The general parsing logic triggers full-document OCR by default. - Phenomenon: Only the content of the default Sheet1 is parsed for imported Excel documents, and quantity data from other worksheets is not extracted. Cause: The rule to traverse and parse by sheet name is not configured. The general parsing module only reads the default worksheet.
- Phenomenon: When importing Word or PPT documents, enabling the PDF enhancement function has no effect, and the parsing result is the same as when the function is not enabled. Cause: The PDF enhancement function only applies to PDF format documents. It has no effect on imports of Word or PPT formats. Parsing configuration was not adjusted for non-PDF documents.
How to confirm the configuration is correct
- Upload a single Excel engineering document with multiple sheets, check if the parsing result includes content from all worksheets, and confirm that the chunking rule adapts to the contextual association of multi-table data.
- Upload a PDF document with embedded images, compare the parsed text with the original document, confirm that no redundant OCR text appears, and verify that the OCR enhancement mode is disabled.
- Upload a single general contracting archiving document with a size over 200 MB, check the upload and parsing status, and confirm that the upload threshold and timeout configuration adapt to the document size.
- Split a long text segment that includes engineering measurement units, check if the chunking result retains the association between measurement units and corresponding project quantities, and verify the appropriateness of the chunk length.
The values listed above are common starting points, and should be evaluated against specific local samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.