What data for this category looks like
Daily documents for this category include sales orders, master production plans, material requirements plans, weekly and daily production schedules, work orders, material shortage lists, delivery change notices, and more. Data sources cover ERP orders and inventory, BOMs and process routes, equipment capacity, staff shifts, work-in-progress, and supplier deliveries. Data update frequencies include daily, weekly, when an order is placed, or when an exception occurs. Document fields include order numbers, material codes, capacity man-hours, work-in-progress quantities, delivery dates, and more. Corresponding units include pieces, hours, days, sets, and others. Some documents include digital signatures and complex nested tables, with significant structural differences.
What constraints these characteristics impose on document parsing and chunking
The wide variety of document types and large structural differences require the parsing engine to adapt to all formats ranging from simple work orders to complex monthly production schedules, to avoid splitting that breaks the associated logic of processes and materials. Frequent data updates and the large size of some documents create clear requirements for parsing timeout limits and maximum file upload sizes. Some documents include digital signatures and complex nested tables, which standard parsing engines cannot fully extract content from. Fields include specific units and associated identifiers. Context associations must be retained during chunking, otherwise subsequent cross-review will fail to match material and capacity information.
How to configure settings
These configurations are optimized for FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to documents with dense fields such as production schedules and work orders, avoiding splitting that breaks the associated relationships between processes and materials |
chunkSplitter | By paragraph + semantic boundary | Adapts to multiple types of production documents, preserving the integrity of order logic and production scheduling rules |
customPdfParse | Enabled | Processes PDF documents with digital signatures and complex nested tables, adapting to various production document formats commonly used by PMC departments |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Adapts to large-volume monthly production schedules and full BOM documents, avoiding mid-task parsing timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports bulk import of large documents such as monthly production plans and full material lists |
chunkSettingMode | Manual configuration | Adjusts chunking parameters for different document types, adapting to the scenario of parallel processing of multiple documents by PMC departments |
Three common misconfigurations
- No valid content returned after uploading a PDF with a digital signature: The
customPdfParseconfiguration is not enabled. The default parsing engine cannot recognize signed PDF formats, resulting in failed content extraction. - A read error occurs when calling the document parsing tool: A locally deployed pdf-marker is not correctly connected to the FastGPT container network, or the
UPLOAD_FILE_MAX_SIZEsetting is smaller than the uploaded file size, causing file transfer interruptions. - Association between material codes and corresponding processes is lost after chunking: The
chunkSizesetting is too small, and complete paragraphs containing associated fields are truncated during splitting, resulting in broken context.
How to confirm the configuration is properly set
- Upload a single PDF of a weekly production schedule with complex nested tables, check if the parsed text retains complete corresponding relationships between processes and material codes.
- Upload a monthly BOM document larger than 100 MB, confirm that the parsing task does not trigger a timeout error, and that the generated chunks have no missing core fields.
- Upload a work order PDF with a digital signature, verify that core business fields such as work order number and delivery date can be extracted after parsing.
- Adjust the
chunkSizeparameter, compare chunking results under different values, and confirm that forced splitting across associated fields does not occur.
The values provided are common starting points and should be evaluated against the reader's own samples.