What this category’s data looks like
Typical users of this category include SRE teams, operations managers, and second-tier technical support. Daily processed documents include shift schedules, handover records, escalation contact lists, and daily shift reports. Data sources cover scheduling systems, fault tickets, monitoring alerts, and organizational contact directories. Update frequency is daily or weekly.
Document structures include fixed fields:
- Shift schedules contain shift time slots, on-duty personnel, and contact information
- Handover records contain pending items and follow-up content
- Escalation contact lists contain positions, contact information, and response paths
- Daily shift reports contain daily faults, processing results, and similar content
What constraints these characteristics impose on document parsing and chunking
Most data sources are structured or semi-structured files exported from systems. Some file formats are non-standard. Parsing must accurately identify field associations to avoid splitting associated information into different chunks.
Update frequency is daily or weekly, so document content is time-sensitive. Parsing and chunking must retain contextual associations. This ensures information within each chunk is complete and aligns with business logic.
Some documents are scanned PDF or Excel files. Support for OCR recognition and column mapping configuration is required.
Shift schedules and handover records have strong content correlation. Chunking must avoid breaking the binding relationship between shifts, personnel, and items. If this binding is broken, subsequent Q&A will fail to accurately match on-duty personnel with corresponding items.
File size and parsing time affect chunking efficiency. Reasonable limits must be set for file upload and parsing duration and size.
How to set configurations
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | The chunked content for shift schedules and handover records must retain the association between shifts, personnel, and items to avoid splitting critical information across different chunks |
chunkSplitter | Semantic chunking | Adapts to the field association characteristics of category documents, retaining business context of shifts, personnel, and items |
customPdfParse | Enable OCR mode | Processes scanned PDF documents such as shift schedules and daily shift reports |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to parsing time for large shift schedule and handover record documents |
UPLOAD_FILE_MAX_SIZE | 500 MB | Limits single-file size to avoid parsing failures caused by overly large files |
chunkSettingMode | Manual configuration | Adapts to the fixed field structure of category documents, enabling precise control of chunking rules |
Three Common Mistakes
- After uploading an Excel-format shift schedule or escalation contact list, the vectorized index is incorrect, with some fields empty. The cause is that Excel column mapping rules are not configured, resulting in failure to identify key fields such as on-duty personnel and contact information during parsing.
- Uploading a scanned PDF-format daily shift report fails to recognize text content, and empty text is returned after parsing. The cause is that OCR mode for
customPdfParseis not enabled, making it impossible to extract text from scanned content. - When invoking the API to add content to knowledge base chunks, the association between handover items and corresponding on-duty personnel is not retained. The cause is that
chunkSizeis set too small, splitting associated information across different chunks.
How to confirm configurations are correct
- Standard-format Excel shift schedules are uploaded. The chunk list parsed by the knowledge base is reviewed, and each chunk is confirmed to contain complete shift time slot, on-duty personnel, and contact information.
- Scanned PDF-format daily shift reports are uploaded. The knowledge base parsing API is invoked, and the returned parsed text is confirmed to contain scanned content and support normal retrieval.
- The API for adding content to knowledge base chunks is invoked. Text containing pending items and on-duty personnel is passed in, and the returned chunked content is confirmed to retain the association between the two.
- The configurations of
UPLOAD_FILE_MAX_SIZEandPARSE_FILE_TIMEOUT_SECONDSare reviewed, and the test file upload size and parsing duration are confirmed to meet the configuration requirements.
The values provided are common starting points and should be measured against local samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.