What this category of data looks like
Declaration materials and file data for building construction general contracting projects mainly originate from stages including project approval, bidding and tendering, construction permit application, site visa, and completion acceptance. The update rhythm follows the progress of individual project cycles, which can last from several months to several years from project initiation to completion archiving. Document structures include fixed-format submission forms, multi-page construction drawings, budget documents, qualification certification files, site visa forms, and more. Some file sets contain mixed content of scanned drawings and editable documents. Fields include project number, floor area (unit: ㎡), project cost (unit: ten thousand yuan), construction unit qualification level, visa date, and other items. Each field follows a clear construction industry standard format.
What constraints these characteristics impose on the document parsing and chunking stage
First, documents contain mixed formats, with scanned drawings and editable documents coexisting. This requires the parsing module to support custom extraction logic adapted to different formats. Second, a single file set may contain dozens to hundreds of pages. The chunk length must balance context integrity and retrieval efficiency: excessive length leads to retrieval redundancy, while insufficient length breaks the association between engineering parameters and their units. Third, fields carry specific industry units (㎡, ten thousand yuan), so chunking must avoid splitting parameters and their corresponding units into different segments. Fourth, document formats vary significantly across different stages, so chunking rules must be adapted separately for submission forms, drawings, and budget documents.
How to configure settings
In FastGPT v4.17.0, the following configuration items can be set in the document parsing module of the knowledge base management interface:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Building project documents contain a large number of engineering parameters with units. This length preserves the complete logic of a single visa or budget document, and avoids splitting parameters and their units |
chunkSplitter | Split by paragraph + header and footer | Building documents have fixed headers marking project names and numbers. Splitting by paragraph preserves the complete context of a single submission document |
customPdfParse | Enabled | Some completion drawings are scanned documents. Custom parsing optimizes the accuracy of image-to-text conversion and adapts to the annotated content of engineering drawings |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large project file sets may exceed 50 pages. The default timeout threshold is insufficient for complete parsing |
UPLOAD_FILE_MAX_SIZE | 1000 MB | A single project's archived file set may contain multiple large drawing PDF files. The upload limit needs to be relaxed to support complete file set uploads |
Three common configuration errors
- No response from the parsing node after uploading a file, with a timeout status displayed on the interface. The cause is failure to adjust the
PARSE_FILE_TIMEOUT_SECONDSparameter. Parsing time for large file sets exceeds the default threshold. - Some scanned drawing PDFs fail to recognize annotated text. The cause is failure to enable the
customPdfParseconfiguration. The default parsing logic cannot adapt to content extraction from scanned engineering drawings. - Segments where "floor area" and "㎡" are separated appear in the chunking results. The cause is an excessively small
chunkSizesetting, which splits engineering parameters and their corresponding units into different chunks.
How to verify correct configuration
- Upload a single 50-page completion drawing PDF, and check whether the parsing status completes within 10 minutes with no timeout errors.
- Upload a file set PDF containing scanned documents, and check whether the parsing result includes text content such as drawing dimensions and materials.
- Randomly select 10 chunking results, and confirm that each chunk contains complete engineering parameters and their corresponding units, with no split breaks.
- Upload an archived file set exceeding 500 MB, and confirm that the upload and parsing processes are not interrupted, with no file size limit errors.
The values listed above are common starting points, and should be evaluated against the reader's own 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.