What this category of data looks like
The main entities for this category of data are development and reform departments, industry authorities including housing and urban-rural development, transportation, and water resources, public resource trading centers, and bidding agencies. Documents handled daily include bidding announcements, pre-qualification documents, bidding documents, bid documents, bid evaluation reports, successful bid candidate notices, and successful bid result announcements. Data sources include national public resource trading platforms, local trading platforms, project approval systems, and bidding and tendering supervision systems. Update frequency is daily. Data fields include specific content such as qualifications, performance, bill of quantities, and construction periods. Structural differences across document types are significant. For example, bid evaluation reports contain independent modules such as expert check-in and bid evaluation records. Bid documents are divided into volumes by commercial bid and technical bid.
What constraints do these characteristics impose on document parsing and chunking
Daily updated data sources require the parsing process to support efficient batch processing, to avoid delays that disrupt subsequent review workflows. Documents of multiple types with large structural differences require parsing and chunking logic to adapt to the inherent module divisions of different documents. For example, accurately split expert records and bid evaluation results in bid evaluation reports, or extract bill of quantities content from bid documents. Precise identification of specific fields such as qualifications and performance requires retaining the binding relationship between fields and their corresponding content during chunking, to avoid losing associated information after splitting. Additionally, errors in this category directly affect project progress and compliance. The accuracy of parsing and chunking directly impacts the efficiency and compliance of subsequent file review. It is necessary to avoid mixing up section information or omitting qualification conditions due to incorrect chunking.
How to set up the configuration
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the common module length of documents in this category, such as individual bid evaluation records in bid evaluation reports or single qualification certificates in bid documents |
chunkSplitter | Split by headings and paragraphs | Aligns with the chapter structure of documents in this category, which are divided by headings, such as bidder qualification requirements in bidding announcements or technical solution chapters in bid documents |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Covers the parsing time required for large bid evaluation reports and bidding documents with multiple attachments in this category, to avoid parsing failures due to timeout |
customPdfParse | Enabled | Supports extracting structured content such as tables and images from documents in this category, such as bill of quantities tables and bid evaluation expert check-in information |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Meets the upload requirements for common large bid document collections and multi-section bidding document packages in this category |
Three common mistakes
- Parsed table data is returned as plain text, with no structured table format. The cause is that the
customPdfParseconfiguration is not enabled, so the parsing module cannot recognize and retain the table structure in the document. - After creating a knowledge base via the API, real-time parsing status information cannot be obtained. The cause is that the knowledge base parsing status query interface provided by FastGPT is not called, and only the return result of the initial creation request is relied upon.
- After uploading a bid document containing images, the explanatory text associated with the images is not correctly extracted. The cause is that a model configuration supporting image recognition is not bound, or the image content extraction function of
customPdfParseis not confirmed to be enabled.
How to confirm the configuration is correct
- Upload a bidding document PDF containing a bill of quantities table, and check if the parsed chunked content retains the table's row-column structure and corresponding values.
- Call the knowledge base creation interface, then query the parsing status via the specified interface, and confirm that three statuses can be returned: parsing in progress, ready, and failed.
- Upload a bid evaluation report PDF containing expert check-in photos, and check if the chunked results include the text description content of the images.
- Upload a multi-section bidding document package larger than 500 MB, and confirm that the upload and parsing processes do not trigger size limit or timeout errors.
The values provided here 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.