Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
Survey and design documents and materials mainly come from deliverables generated during the full project cycle of survey and design, with update rhythm following project phase progression. Common document formats include PDFs with embedded charts, Word files with structured tables, and Excel calculation spreadsheets. Document structures include project identifiers, technical parameter descriptions, calculation processes, and attached drawings and tables. Fields include project numbers, survey point coordinates, material strength grades, design load values, and units such as meters, megapascals, cubic meters, and kilonewtons. Some documents include vector charts exported from CAD, which are bound to text content to present technical details.
What constraints these characteristics impose on the document parsing and chunking process
Survey and design documents contain a large number of technical charts and structured calculation tables. Conventional text parsing cannot fully extract visual content and structured data. Document chapter levels are clear, but paragraph lengths vary greatly. Short parameter descriptions are mixed with long calculation processes, so chunking must balance semantic completeness and retrieval granularity. Some documents embed CAD-exported vector images, so parsing must support extraction of non-bitmap image formats. Metadata is scattered across document headers and chapter notes, and must be bound to corresponding text blocks to ensure accurate subsequent retrieval. Parsing large documents such as full construction drawing collections requires sufficient time and storage space to avoid process interruptions.
How to configure the settings
| Configuration Item | Suggested Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the average length of technical paragraphs in survey and design documents, avoiding splitting complete calculation processes or parameter descriptions |
chunkSplitter | Split by paragraph + title boundaries | Aligns with the document's chapter hierarchy, retains semantic integrity of technical content, and avoids breaking logical connections within paragraphs |
customPdfParse | Enabled | Supports extraction of embedded vector charts and structured text from PDFs, resolving the issue of unviewable PDF images |
UPLOAD_FILE_MAX_SIZE | 500 MB | Meets upload requirements for full construction drawing collections and large survey reports, preventing blocking due to oversized files |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Prevents timeout during large document parsing, ensuring smooth completion of full content extraction workflows |
chunkSettingMode | Manual mode | Allows adjustment of chunking parameters for different document types such as survey reports and construction drawing collections, to adapt to diverse document structures |
Three Common Configuration Errors
- Phenomenon: Embedded charts cannot be displayed in the interface after uploading a PDF, and the parsing result only contains plain text. Cause: The
customPdfParseconfiguration is not enabled, and the default parsing engine cannot extract technical charts in vector format. - Phenomenon: Search results only return text blocks without associated metadata such as project numbers and survey point coordinates. Cause: Context paragraphs containing metadata are not retained during chunking, or metadata is not bound to corresponding text blocks.
- Phenomenon: Parsing fails when uploading Excel-format survey calculation spreadsheets. Cause: Parsing rules adapted to Excel structured tables are not configured, or the chunking logic does not distinguish between tables and text content.
How to Verify Correct Configuration
- Upload a PDF survey report containing embedded vector charts, and check if the parsing result includes chart preview content.
- Upload a large construction drawing collection under 500 MB, and confirm that the upload and parsing processes have no timeout errors or interruptions.
- Search for keywords of project numbers in the document, and confirm that the returned text blocks are accompanied by corresponding project identifier metadata.
- Upload an Excel-format calculation spreadsheet, and confirm that the parsing result retains the structured content and original numerical values of the table.
The values provided are common starting points and should be evaluated against samples relevant to the specific workflow.
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.