Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
Survey and design business data originates primarily from survey and mapping reports, construction drawing design documents, submission and approval documents, and on-site visa materials generated after project approval. The data update rhythm follows project phase progression, with core version updates occurring once during the preliminary design phase, construction drawing phase, and completion archiving phase. Document structures include fixed chapter frameworks, such as project overview, geotechnical parameters, design plans, calculation reports, and approval opinions. These often contain nested tables and professional units, such as geotechnical bearing capacity characteristic values (kPa), building height (m), and building floor area (㎡). Some documents are exported from CAD, with format features including vector text and nested tables.
What constraints do these features impose on the document parsing and chunking step
The professional chapter structure and nested table features of survey and design documents require chunking to preserve the binding relationship between chapter titles and core content, to avoid disrupting professional logic. Image-format content in scanned survey reports requires the parsing link to support OCR recognition to extract valid text. Parsing duration requirements are higher for large multi-page documents, so support for long processing is needed. The association between professional units and parameters must be retained in the same chunk, to avoid losing professional context after splitting, which affects the accuracy of subsequent business process guidance.
How to configure settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Survey and design documents include long paragraphs of calculation reports and professional tables. This range preserves the complete logic of a single professional module. |
chunkSplitter | Split by paragraph + title anchor | Documents have clear chapter titles and professional paragraphs. Anchor splitting avoids cross-chapter content concatenation. |
customPdfParse | Enable structured parsing (including scanned OCR adaptation) | Survey and design PDFs often contain nested tables and scanned content. Enabling this preserves table field associations and recognizes text in images. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large survey report PDFs have many pages, requiring sufficient time to complete the parsing and chunking process. |
chunkSettingMode | Manual configuration | Adjustment of chunking parameters is required for different types of survey and design documents (such as survey reports, construction drawing descriptions). Automatic mode cannot adapt to professional scenarios. |
Three common mistakes
- Search results return irrelevant content, and target geotechnical parameters or design plan paragraphs cannot be located. The cause is that the association between chapter titles and core content was not preserved during chunking, and cross-chapter professional content was split into different chunks, leading to failure to match accurate context during retrieval.
- After uploading a scanned survey report PDF, the parsed content only contains a small amount of garbled text or no valid text. The cause is that structured parsing configuration for scanned documents was not enabled. General parsing can only recognize vector text and cannot extract professional content from images.
- Uploading a large construction drawing description document triggers parsing timeout, and the interface shows parsing failure. The cause is that
PARSE_FILE_TIMEOUT_SECONDSuses the default short duration, which cannot complete the complete parsing process for multi-page complex documents.
How to confirm correct configuration
- Upload a typical survey report PDF, check the parsed chunk preview, and confirm that each chunk contains a complete professional paragraph and associated chapter title.
- Enter a professional search term (such as "geotechnical bearing capacity characteristic values") for a search test, and confirm that the returned results include the target content and have no irrelevant cross-chapter paragraphs.
- Upload a scanned survey document, and confirm that the parsing result can extract professional tables and text content from the image.
- Adjust the
chunkSizeparameter, compare the chunking effects under different values, and confirm that it meets the professional logic splitting requirements of the current document.
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.