What the data for this category looks like
The business data for survey and design mainly comes from project survey reports, design disclosure documents, quantity takeoff ledgers, progress weekly reports, and other sources. The update rhythm is dynamically adjusted based on project milestones, with no fixed cycle. Document formats include plain text, structured tables, and annotated PDF drawing instructions. Fields cover project numbers, survey point coordinates, material usage, duration milestones, and more. Units include meters, cubic meters, man-days, ten thousand yuan, and others. Some documents contain mixed text and images, or combine scanned documents with text annotations.
What constraints do these characteristics impose on the document parsing and chunking workflow
A large number of structured tables are mixed with long text descriptions in survey and design documents. When chunking documents, field associations must be retained, and splitting table cells must be avoided to prevent semantic discontinuity. Document lengths vary significantly across different projects, ranging from a few pages of point records to dozens of pages of general reports. Adaptive chunking parameter adjustment is required. Some documents contain scanned materials and handwritten annotations. Enhanced parsing capabilities are needed to extract non-standard text content. Batch uploaded ledger data must maintain chunking consistency, and segmentation differences across different upload channels must be avoided.
How to Configure Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapt to the length of long text descriptions and structured tables in survey and design documents, retain complete semantics of a single ledger entry or single segment of survey content |
chunkSplitter | Split by paragraph + table boundaries | Avoid splitting table cells, retain contextual association of structured data |
customPdfParse | Enabled | Adapt to common scanned annotations and mixed text-image documents, improve parsing completeness of non-standard format content |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Reserve sufficient time to parse large survey reports, avoid parsing interruptions caused by timeouts |
chunkSettingMode | Manual configuration | Flexibly adjust chunking parameters for different types of survey and design documents such as point records and general reports |
Three Common Configuration Mistakes
- Phenomenon: Chunking results for the same file differ between API upload and platform direct upload. Cause: The
chunkSplitterparameter was not specified during API upload, so the default chunking rule was used, which differs from the default configuration for platform direct upload. - Phenomenon: No parsed data is available for documents uploaded via custom links after enabling
customPdfParse. Cause: The custom link does not have open access permissions, or the document format exceeds the range supported by enhanced parsing. - Phenomenon: After configuring the document parsing node, the model cannot read the uploaded file. Cause:
PARSE_FILE_TIMEOUT_SECONDSis set too short, so large survey reports are terminated before parsing completes, orchunkSizeis set too large, causing chunks to exceed context limits.
How to Confirm Configuration Is Complete
- A single small survey ledger Excel file is uploaded. The parsed chunking results are reviewed to confirm that table cells are not split.
- A single large PDF survey report is uploaded. Parsing progress is verified to complete within the time set by
PARSE_FILE_TIMEOUT_SECONDS. - The same document is uploaded via API and platform direct upload separately. Chunking results are compared to confirm consistency.
- After enabling
customPdfParse, a document with scanned annotations is uploaded. Annotation content is confirmed to be correctly extracted.
The values provided are common starting points and should be measured 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.