This document is written based on FastGPT v4.17.0.
What this category of data looks like
Documents used daily by PMC, planning, shop scheduling, and material control departments of various manufacturing enterprises include sales orders, master production plans, material requirement plans, weekly production scheduling tables, work orders, material shortage lists, and delivery change notices. Data sources include ERP orders and inventory, BOM and process routes, equipment capacity, staff shifts, work in progress, and supplier arrivals. Update frequency is daily, weekly, when an order is created, or when an exception occurs. Documents contain structured tables and unstructured text, with fields including order numbers, material codes, capacity units, scheduling node times, material usage, and more.
What constraints do these characteristics impose on the document parsing and chunking link
Documents in this category contain both structured tables and unstructured text. Chunking rules must balance the integrity of table rows and the coherence of text semantics. They must avoid splitting associated fields. Frequently updated business data requires parsed chunks to quickly match the latest scheduling and material information. Chunks that are too large lead to redundancy. Chunks that are too small cause broken associations. Specific business fields and units require retaining contextual connections during chunking. This ensures accurate matching of orders, schedules, and corresponding resources during retrieval. Documents for tender scenarios must accurately correspond to delivery commitments. Chunks must be split by business unit. They must avoid mixing content across work orders or processes.
How to set the configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | This category of documents includes structured scheduling tables and long-text change notices. This range preserves complete associations between core fields such as processes and capacity |
chunkSplitter | Semantic + header splitting | Scheduling tables in this category of documents have fixed headers. Splitting by semantics avoids splitting table rows and their corresponding explanatory text |
customPdfParse | Enable structured extraction | PDFs in this category are mostly fixed-layout production reports or scanned documents. Enabling this preserves table fields and structural information |
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | Large multi-work order or BOM documents require longer processing time. This setting avoids interrupting complete parsing due to timeout |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Bulk multi-work order documents in this category are usually large in size. This setting prevents upload failures due to file size limits |
chunkSettingMode | Manual configuration | Document formats vary widely across this category. Manual configuration allows flexible adjustment of chunking rules to adapt to different types of business documents |
Three common mistakes
- Search result matching deviation or no results: Core associated fields such as order numbers and material codes are not retained during chunking. This prevents retrieval from linking to corresponding scheduling and material data.
- Missing fields after parsing scanned production reports: Structured scanning recognition mode for
customPdfParseis not enabled. Only unstructured image text is extracted, and table fields cannot be restored. - Parsing task timeout interruption: The value set for
PARSE_FILE_TIMEOUT_SECONDSis lower than the parsing time required for large multi-work order documents in this category. This causes tasks to terminate early.
How to confirm proper configuration
- Upload a mixed document containing sales orders and weekly production scheduling tables. View the chunk preview. Confirm that each chunk contains complete order numbers, corresponding processes, and capacity data.
- Upload a scanned PDF of a material shortage list. Check if the parsed text retains table fields such as material codes, shortage quantities, and supplier information, with no obvious missing content.
- Test uploading 10 bulk work order documents. Confirm that all parsing tasks complete within the set
PARSE_FILE_TIMEOUT_SECONDSvalue, with no timeout errors. - Initiate a retrieval test. Enter scheduling keywords for a corresponding order. Confirm that the returned chunks contain the complete scheduling content for that order.
The values provided are common starting points and should be measured against the reader's own samples.