What the data for this use case looks like
Shipbuilding ledger data primarily comes from sectional construction ledgers, material requisition records, hull welding inspection reports, main engine assembly archives, and similar documents. Sources include docx files exported from project management systems, scanned PDFs from production sites, and Confluence pages on internal network project wikis.
Data update rhythm follows construction milestones: Ledger documents for a single sectional are updated after assembly is completed. Full ship ledgers are summarized on a quarterly basis. Most documents include fixed fields: sectional number, material code, requisition quantity, inspector ID, and operation duration. Units cover engineering-related categories such as meters, kilograms, hours, and pieces.
What constraints do these characteristics impose on the document parsing and chunking workflow?
The combined characteristics of fixed fields and engineering units in shipbuilding ledgers require that context links between fields and their corresponding values be preserved during document parsing and chunking. This prevents cases where fields become separated from their associated values after chunking.
Documents from multiple sources include structured docx files, scanned PDFs, and internal network Confluence pages. Different parsing logic must be adapted for each format. OCR functionality must be enabled for scanned PDFs.
Single full ship summary documents have significant length. Chunking must avoid truncating continuous content from key operation processes.
For internal deployment scenarios, the internal network access permission issue for Confluence pages must be resolved to ensure normal content extraction from internal documents.
How to configure the settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Basis for This Value |
|---|---|---|
chunkSize | 800–1200 characters | Shipbuilding ledgers contain multiple sets of fields and values. This range preserves complete context for a single ledger entry or continuous operation records |
chunkSplitter | Split by paragraph + fixed fields | Ledger documents typically separate operation records for different sections using paragraphs or fixed headers. This avoids splicing content across sections |
customPdfParse | Enabled | PDFs for shipbuilding ledgers often include scanned documents or engineering drawing attachments with complex formatting. Custom parsing optimizes OCR recognition accuracy and formatting restoration |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Single full ship summary documents have significant length. Sufficient time must be reserved for parsing and OCR processing |
chunkSettingMode | Manual configuration | Chunking rules must be adjusted based on the fixed field structure of ledgers. This prevents automatic chunking from breaking field associations |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Shipbuilding ledger documents may include multiple drawing attachments. This setting supports batch upload of large files |
Three Common Misconfigurations
- Issue: Parsed ledger content is merged into a single paragraph with no structured formatting. Cause:
customPdfParseis not enabled, or chunking rules do not use paragraph-based splitting. This results in parsed content failing to retain the original formatting structure. - Issue: Batch uploads of multiple ledger documents cause parsing tasks to time out and fail. Cause:
PARSE_FILE_TIMEOUT_SECONDSis not adjusted to a value suitable for long documents, or the number of parallel uploads exceeds the system default limit. - Issue: Content cannot be extracted from internal network Confluence pages. Cause: Internal network access permissions are not configured, or internal network parsing adaptation is not enabled. This prevents the system from accessing internal wiki pages.
How to Confirm Proper Configuration
- Upload a typical ship sectional ledger docx document, and verify that the parsed content retains the corresponding relationship between fields such as sectional number and material code.
- Upload a scanned inspection report PDF, and confirm that the parsed text includes OCR-identified inspector signature and date fields.
- Upload a lengthy full ship summary document, and check that chunked content does not truncate continuous operation record paragraphs.
- After configuring the access address for internal network Confluence, attempt to import internal wiki pages, and confirm that page content can be parsed normally.
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.