What the Data for This Use Case Looks Like
Data sources include the national medical insurance information platform, hospital HIS and medical insurance interfaces, medical insurance catalog coding database, and cross-regional medical settlement platform. Update frequency is daily. Documents processed include medical settlement lists, expense details, medical record homepages, prescriptions, rejection lists, and settlement statements. Fields include diagnosis and treatment item codes, pharmaceutical and consumable codes, expense amounts, insured person information, settlement status, and more. All fields comply with national unified medical insurance standards, with no additional custom units.
Constraints on Knowledge Base Retrieval and Recall
Scattered data sources requiring integration with multiple system interfaces mandate that the retrieval pipeline support unified coding matching across cross-source data. Daily updated medical insurance catalog and settlement data require retrieval to support high-frequency incremental synchronization to maintain data timeliness. Documents include both structured coding fields and unstructured diagnosis descriptions. This requires retrieval to accommodate both precise coding matching and semantic understanding needs. Audit scenarios for single or bulk settlements require retrieval to flexibly adjust recall scale to match different business volumes. Rejection reconciliation processing requires recall results to link complete expense and prescription related information, preventing data disconnection.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for medical insurance coding and expense details. A setting that is too low will introduce irrelevant settlement data. A setting that is too high may miss valid matching entries. |
limit | top 10 entries | Fits the single recall scale for single or bulk settlement audits, avoiding excessive redundant data that slows down the audit process. |
searchMode | semantic retrieval | Medical insurance documents contain a large number of unstructured diagnosis descriptions and expense explanations. Semantic retrieval can more accurately match scenario descriptions in user audit queries. |
usingReRank | enabled | Reranks recalled settlement and catalog data, prioritizing returning diagnosis items and expense entries most relevant to audit requirements. |
datasetSearchUsingExtensionQuery | enabled | Automatically expands user audit queries, supplementing default fields such as medical insurance coding and settlement status, improving recall precision. |
collectionIds | bind medical insurance settlement database, medical insurance catalog database, cross-regional medical settlement database | Covers all data sources required for this use case, preventing retrieval from missing key documents. |
Three Common Misconfigurations
- Phenomenon: Retrieval results appear intermittently, and some audit requests do not trigger knowledge base recall. Cause: Forced retrieval trigger rules are not configured. Retrieval is only called when the system determines it is necessary by default, leading to some coding matching requests missing knowledge base calls.
- Phenomenon: Retrieval latency increases significantly when consecutive audit requests are initiated in the same session, and some requests time out after 60 seconds. Cause: Concurrent retrieval limits within the session are not set. Multiple requests reuse the same knowledge base index loading resources, leading to excessive resource occupancy.
- Phenomenon: The accuracy of rejection list matching results returned under mixed retrieval mode is low. Cause: The retrieval mode is not adjusted to prioritize coding matching for medical insurance settlements. Keyword matching in mixed retrieval cannot fully cover structured coding fields of medical insurance data.
How to Verify Proper Configuration
- Initiate a test request for single medical insurance expense settlement audit, check if retrieval results include medical insurance catalog coding information for the corresponding diagnosis item.
- Initiate 10 consecutive bulk audit requests, observe if retrieval latency remains stable with no significant fluctuations.
- Switch retrieval mode to mixed retrieval, compare the coding matching accuracy of returned results, confirm that semantic retrieval mode performance meets business requirements.
- View knowledge base synchronization logs, confirm that daily incremental update tasks have executed normally with no data synchronization failure records.
The configuration values listed above are common starting points. They should be evaluated against the specific use case’s own sample data.
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.