What this category of data looks like
Data for this category comes primarily from the national medical security information platform, hospital HIS and medical insurance interfaces, medical insurance catalog coding database, and cross-regional medical settlement platform. Daily processed documents include medical insurance settlement lists, expense details, medical record homepages, prescriptions, rejection lists, and settlement reconciliation statements. Data updates daily. Each individual document contains fields such as diagnosis and treatment item codes, drug and consumable codes, expense amount, insured person identifier, settlement time, and rejection reason. No unified fixed format exists, but core fields are code-related and amount-related.
Constraints on Vector Models and Indexing
Data updates daily and comes from multiple dispersed sources. Incremental index construction and unified indexing for multi-source heterogeneous data must be supported. Core fields include structured text such as diagnosis and treatment, drug and consumable codes. Chunking and vector generation logic must be optimized for code-related content. Individual review tasks require associating multiple settlement and prescription documents. Directional recall across knowledge bases must be supported. The scale of documents processed in a single task is large. Index loading and retrieval response times must be controlled to avoid impacting daily review efficiency.
How to Set Configurations
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Matches the per-page content length of documents such as settlement lists and prescriptions, avoids overly fragmented chunks or splitting across critical information |
| Overlap Length | 100–150 characters | Covers cross-chunk associations of core fields such as codes and amounts, ensures critical matching items are not lost during retrieval |
| Similarity Threshold | 0.75–0.85 | Meets the precision requirements for matching diagnosis and treatment codes, avoids mismatches that lead to review errors |
| Single Search Limit | Top 10 results | Balances recall efficiency and the amount of associated information required for reviews, covers common multi-document association scenarios |
defaultToken | 4096 | Adapts to the context length requirements of single-batch review tasks, including multiple chunks of settlement details, prescriptions, rejection records, and other content |
maxToken | 8192 | Reserves sufficient context space to handle long-text processing scenarios for batch settlement reviews |
Three Common Misconfigurations
- Retrieval results include content from two knowledge bases simultaneously, and do not prioritize recall of specified knowledge base content per preset priority. This occurs because knowledge base priority binding is not enabled in the vector model retrieval configuration, or weight parameters for the corresponding knowledge base are not configured.
- Retrieval tasks return a 60-second timeout error. This occurs because the single search limit is set too high, and incremental recall logic is not optimized for daily updated data indexes, leading to excessive full index loading time.
- Bulk index addition requests return parameter validation failures. This occurs because per-item data parameters for bulk upload are not set according to the maximum length requirements of collection chunks, causing some document chunks to exceed the allowed range.
How to Confirm Configurations Are Correctly Set
- Upload a single medical insurance settlement list document, run a retrieval test, and check whether recall results prioritize the knowledge base with the highest preset priority.
- Submit a retrieval task for batch settlement reviews, view task duration, and adjust configuration items until they meet business response requirements.
- Run a bulk index addition test request, and check whether returned results have no parameter error prompts.
- View index rebuild status, confirm that incremental indexes for daily updated data complete updates within a reasonable time frame.
The values provided here are common starting points, and should be evaluated against the reader's own sample data.
Reference only; verify against the target FastGPT release and deployment documentation.
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.