What data for this category looks like
Data related to shipbuilding service inquiries and replies originates primarily from PDF ship design drawings, construction process manuals, classification society inspection reports, project approval application materials, and archived past inquiry replies. Data updates are triggered by the launch of new shipbuilding projects or annual maintenance plans, and iterated on demand during project cycles. Most documents are long-form structured files, with clear section titles, technical parameter tables, and embedded CAD drawings. Fields include ship type number, material specification, inspection date, approval document number, and more. Units use standard engineering units such as millimeters, tons, and kilowatts.
What constraints these characteristics impose on the document parsing and chunking link
The long-form and structured nature of shipbuilding documents requires chunking to retain associations between sections and parameters, to avoid breaking the integrity of process descriptions. Embedded CAD drawings and tables increase parsing complexity. Without targeted parsing rules, structured data may be lost. Project document volumes vary widely, ranging from single inspection reports to full sets of construction manuals. Parsing time limits must adapt to these different volumes. At the same time, metadata for service inquiries such as approval document numbers must be bound to corresponding content. Chunking must retain metadata associations, otherwise recall cannot accurately match inquiry needs.
How to configure settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Shipbuilding documents contain long process paragraphs and parameter tables. This range retains the contextual integrity of a single parameter group or complete process paragraph, avoiding chunk breakage |
chunkSplitter | Split by paragraph + title | Shipbuilding documents have clear section hierarchies and process paragraph divisions. Splitting by this rule retains document structural relevance, facilitating subsequent recall matching of specific inquiry links |
customPdfParse | Enabled | Shipbuilding documents include embedded CAD drawings and structured tables. Enabling this setting fully extracts text data from tables and attached drawings, avoiding loss of key technical information |
UPLOAD_FILE_MAX_SIZE | 1000 MB | A full set of construction manuals may exceed 500 MB. This value covers the upload needs of most project documents |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing large, multi-page PDF documents requires longer processing time. This duration prevents the parsing process from being interrupted mid-flow due to timeout |
chunkSettingMode | Manual configuration | Shipbuilding document types vary significantly. Manual configuration allows flexible adjustment of chunking rules to adapt to splitting needs for drawing-based, manual-based, and other document types |
Three common mistakes
- Calling the parsing API returns the error
Cannot read properties of undefined. The cause is thatcustomPdfParseis not enabled, and the document contains embedded tables. This causes the parser to fail to recognize table structures and return empty fields. - The proportion of duplicate content after chunking is too high. The cause is that no reasonable
chunkSplitterrule is set. Adjacent duplicate parameter descriptions are split into multiple independent chunks, leading to repeated hits of the same content during recall. - Parsing fails when uploading large-volume documents. The cause is that the set
PARSE_FILE_TIMEOUT_SECONDSvalue is less than the actual required parsing time, causing the parsing process to be forcibly interrupted before completion.
How to confirm the configuration is correct
- Upload a single typical ship inspection report. Check the parsed text content to confirm that embedded inspection tables and parameters are fully extracted.
- Adjust the value range of
chunkSize. Compare chunking results under different configurations to confirm that chunks retain the association between a single set of technical parameters or a complete process paragraph. - Upload a preset maximum-volume project document. Confirm that the parsing process does not trigger a timeout interruption, and that complete chunking results are generated.
- Simulate a service inquiry scenario. Enter a query containing ship type parameters. Confirm that the recalled chunks can accurately associate corresponding document sections and technical data.
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.