What data falls under this category
The data for this category mainly comes from the medical insurance intelligent monitoring system, fund supervision rule base, medical insurance settlement data, drug and consumable traceability codes, and complaint clues. Commonly processed documents include audit notices, inspection work papers, list of suspected issues, interview records, decision documents, and rectification reports. Data is updated on a monthly basis. Core documents often contain a mix of structured fields and unstructured text. They require linking to multi-source data such as settlement details and traceability information. Some documents also include embedded tables and image attachments.
What constraints do these characteristics impose on the document parsing and chunking link?
Multi-source mixed data structures require the parsing process to adapt to both structured tables and non-standard text. Monthly batch documents require stable parsing throughput. For documents linked to multi-source data such as settlement and traceability information, context association must be retained during chunking. This avoids splitting cross-field content related to suspected issues. Audit documents have strict compliance requirements. Parsing results must not omit key information such as inspectors and the basis for violation determination. Omitting this information will hinder the formation of follow-up handling opinions. For documents with embedded tables and images, precise text extraction from these elements is required. This prevents loss of key information.
How to configure settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Documents in this category mostly have mixed structures. This range retains key context such as suspected issue descriptions and handling basis. It adapts to the chunking logic of FastGPT v4.17.0 |
chunkSplitter | By paragraph + semantic boundary | Audit documents often have clear inspection stages and suspected issue sections. This avoids splitting logically related content |
customPdfParse | Enabled | Daily documents are mostly in PDF format. Enabling this option optimizes the parsing effect of tables and embedded images |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Monthly batch documents have large individual file sizes. Sufficient time must be reserved for parsing and OCR processing |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Some audit documents are linked to multiple settlement detail attachments. Bulk upload of large files must be supported |
Three common configuration errors
- After uploading image files or PDF files with embedded images, the parsing result contains no text related to the images. This occurs when the embedded image OCR parsing switch for
customPdfParseis not enabled, or when call permission for the corresponding OCR model is not configured. - After importing an Excel knowledge base backup packaged from an old server into a new server, the chunking and vector association configurations of the original knowledge base are not fully restored. This occurs when the official FastGPT bulk export and import tool is not used, and manually packaged files do not retain system-level configuration information.
- The table dataset type cannot be selected during the knowledge base creation process. This occurs when the permission switch for the table parsing module is not enabled, or when the current version's data source configuration does not cover the table parsing function.
How to confirm the configuration is correct
- Upload a single typical audit work paper. Review the parsed chunk list, and confirm that key fields such as violation determination basis and inspectors are not split across chunks.
- Upload an audit report with embedded tables and images. Verify that the parsing result includes the detailed data in the tables and the annotation text in the images.
- Adjust the
chunkSizeparameter. Re-parse the same document, compare the number of chunks and content integrity, and confirm that it meets business requirements. - Test bulk upload of monthly batch documents. Check the completion status and error rate of parsing tasks, and confirm that it meets batch processing requirements.
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.