Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
Shipbuilding business document sources include PDFs exported from hull design CAD, Javadoc-format interface documents from equipment manufacturers, PDFs converted from ship classification society inspection scans, production scheduling spreadsheets, and more. Document update cadence adjusts based on construction phases: design documents are updated monthly as projects progress, equipment interface documents are updated quarterly with hardware iterations, and inspection reports are generated immediately per production batch. Document structures include long chapter manuals, nested HTML interface documents, and annotated drawing PDFs. Fields include ship classification society numbers, steel plate grades, weld numbers, torque values, and come with engineering-specific units such as mm, N·m.
What constraints these characteristics impose on document parsing and chunking
Long documents such as hull design manuals can span hundreds of pages. Generic chunking rules may split critical engineering parameters from their surrounding context. Javadoc-format interface documents contain multiple nested tags and structured comments; default parsers cannot fully extract core content like interface methods and parameter descriptions. If fields with engineering units such as 12mm thickness are split, subsequent semantic disconnection during retrieval will occur. Scanned inspection report PDFs require OCR to extract text from images, which standard text parsing cannot cover.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the paragraph length of long shipbuilding documents, avoids splitting core engineering parameters from their context |
chunkSplitter | Split by paragraph + heading | Matches the chapter hierarchy of shipbuilding documents, preserves logical units of inspection reports and process manuals |
customPdfParse | Enable OCR mode | Processes scanned ship classification society inspection report PDFs, extracts text embedded in images |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Meets upload requirements for large documents such as final assembly process manuals, prevents file blocking |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Reserves sufficient time for long document parsing, prevents mid-parsing timeout interruptions |
chunkSettingMode | Manual configuration | Flexibly adjusts chunking rules for special fields and structures of shipbuilding documents |
Three Common Mistakes
- After uploading HTML interface documents generated from Javadoc, the knowledge base has no parsed content. The cause is that parsing rules adapted to HTML format are not enabled; the default parser only recognizes standard PDF and plain text formats.
- After uploading a PDF document containing inspection data images, text content from the images cannot be extracted. The cause is that OCR mode for
customPdfParseis not enabled; the default parser only extracts embedded text in PDFs. - Chunking results split steel plate grades and their units, such as "12mm thickness" split into separate segments. The cause is that
chunkSplitteris only set to split by character length, not configured to split by semantic units.
How to Confirm Proper Configuration
- Upload a single Javadoc-format equipment interface document, check if the parsed text content includes structured information such as interface methods and parameter descriptions, confirm that core content is correctly extracted.
- Upload a PDF document containing inspection data images, check if the parsed text includes text information from the images, confirm that OCR extraction is active.
- Upload a scanned inspection report PDF with more than 100 pages, check if the parsing progress completes within the time set by
PARSE_FILE_TIMEOUT_SECONDS, with no timeout errors. - Randomly select a text segment containing engineering units, check if the chunking results fully retain the combination of parameters and units, with no semantic splitting.
- Check the file upload logs in the knowledge base, confirm that large documents were not blocked due to exceeding
UPLOAD_FILE_MAX_SIZE, and successfully entered the parsing process.
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.