Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this use case looks like
This use case supports internal office audit workflows and employee business support for entities including medical insurance handling agencies and designated hospital medical insurance offices. Data is sourced from the national medical security information platform, hospital HIS and medical insurance interfaces, medical insurance catalog coding database, and cross-regional medical settlement platform. The update frequency is daily.
Commonly processed documents include medical insurance settlement lists, expense details, medical record homepages, prescriptions, rejection lists, and settlement reconciliation statements. These documents contain fields such as settlement number, diagnosis and treatment project code, expense amount, insured person identifier, and rejection reason. Some documents use structured formats, while others are scanned or exported unstructured files.
Constraints Imposed on Document Parsing and Chunking
The daily batch-updated document scale requires parsing tools to have stable batch processing capabilities, to match the high-frequency audit rhythm of internal office work. Documents include structured coding fields and unstructured explanatory text, so support must be provided for both structured data extraction and general text parsing. This avoids losing key coding information required for internal audits.
Individual settlement-related documents contain multiple sets of business entries. Chunking must retain the contextual association between entries, to allow internal employees to quickly match expenses with corresponding diagnosis and treatment projects. Rejection lists contain dispute reasons and verification information. Chunking must retain the logical order of entries, to simplify the appeal and reconciliation process for internal employees.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | A single set of business entries in medical insurance settlement details and prescriptions includes codes and expense descriptions. 800–1200 characters can retain the complete context of a single set of entries, allowing internal employees to quickly match expenses with corresponding diagnosis and treatment projects |
chunkSplitter | Split by punctuation and semantic meaning | Medical insurance documents include structured codes and natural language descriptions. Splitting by semantic meaning avoids separating codes from their corresponding expense descriptions, adapting to the information association needs of internal audits |
customPdfParse | Enabled | Some medical insurance settlement documents use customized PDF formats. Enabling custom parsing can accurately extract structured fields, avoiding loss of key coding information required for internal audits via general parsing |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Batch-processed settlement reconciliation statements may include multiple batches of data. 1000 MB covers the size upper limit of conventional batch documents, adapting to the batch processing needs of internal office work |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing large batch documents takes a long time. 600 seconds covers the parsing cycle of conventional documents, avoiding timeout interruptions that disrupt internal office rhythms |
chunkSettingMode | Manual configuration | The fields and structure of medical insurance documents are fixed. Manual configuration can precisely match the business needs of internal audits, avoiding unreasonable splitting from automatic chunking |
Common Configuration Mistakes
- Uploading a link results in failure to parse medical insurance document content, only returning empty content or timeout prompts. This occurs when the
customPdfParseconfiguration is not enabled, or when parsing rules for customized PDF or web formats of medical insurance documents are not adapted. - Diagnosis and treatment project codes and their corresponding expense descriptions are split into different chunks in the chunking result. This occurs when
chunkSizeis set too small, or whenchunkSplitteris not configured to split by semantic meaning, leading to separation of business association information and reducing the audit efficiency of internal employees. - Knowledge base chunking displays normally, but context references do not retain the original document's format structure. This occurs when
customPdfParseis not configured to extract formats from structured documents, or when original document format markers are lost during chunking, reducing the readability of context references.
How to Verify Correct Configuration
- Upload a single medical insurance settlement list, check whether the parsed text completely extracts key fields such as settlement number, diagnosis and treatment project code, and expense amount.
- Batch upload the settlement detail documents updated on the current day, check whether the completion rate and time consumption of parsing tasks match the update rhythm of internal office work.
- Review the chunking results, confirm that each chunk contains complete business entry association information, with no key fields split across chunks.
- Verify the chunk ID copy function, confirm that the identification information of the corresponding chunk can be copied normally, to allow internal employees to quickly locate relevant documents.
The values provided are common starting points and should be evaluated 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.