What the data for this category looks like
Core data for this category comes from design history files, device and material batch ledgers, MES systems, clean environment monitoring equipment, sterilization equipment, and inspection and calibration systems. Daily generated documents include production process specifications, batch production records, clean zone temperature, humidity and differential pressure records, sterilization batch records, third-party inspection reports, software version logs, adverse event and traceability materials. Data updates occur once per batch or per device, with clean environment monitoring data updated per the established monitoring plan. Document fields include unique device identification (UDI), batch number, sterilization parameters, clean zone indicators, inspection results, and more. Units include ℃, Pa, minutes, batch number strings, and others.
Constraints imposed by these characteristics on source citation and traceability
Dispersed multi-source data requires association of batch ledger information from different systems, which places requirements on the accuracy of cross-source recall. Frequently updated batch records and monitoring data require traceability processes to support near-real-time document retrieval and updates. Diverse document structures include structured batch ledgers and unstructured process specifications, which require adaptive content splitting and matching for different formats. Precise matching requirements for key fields such as UDI and batch number require retrieval logic to prioritize association with core identification fields. Severe error consequences require traceability results to fully trace back to original documents and data sources, to avoid incorrect or missed citations.
How to set up configurations
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Single search limit | Top 6 entries | Covers core documents related to materials, sterilization, inspection, and more involved in single-batch medical device production, meeting multi-document association requirements for supplier audits |
| Maximum context | 12000–15000 characters | Accommodates complete fragments of multiple associated batch records and clean zone monitoring data, supporting complete audit logic |
| Reranking | Enabled | Prioritizes content matching key traceability fields such as UDI and batch number, improving accurate recall rate |
quoteMaxToken | 8000 token | Ensures recalled document fragments are sufficient to support traceability and audit logic, avoiding information loss due to content truncation |
responseData | Return associated batch ID and document source path | Meets compliance traceability requirements, facilitating audit personnel to verify original data |
detail | Enabled | Displays specific fields and values of document fragments, facilitating audit personnel to quickly locate key audit points |
Three common mistakes
- When initiating supplier audit queries in simple workspace knowledge base applications, only 1 document is returned per query, making it impossible to associate multiple batch records and inspection reports. The cause is an overly low single search limit configuration that does not cover multi-document association scenarios involved in single-batch production.
- After importing a preset JSON-format knowledge base, there is no default option in the knowledge base selection drop-down menu, and the association path must be manually configured. The cause is failure to add metadata tags and default association rules to the imported knowledge base.
- An error is prompted when testing the model after configuring
responseData, but normal citation operation can be performed after forcibly ignoring it. The cause is that the context length of the test environment does not match the formal operating configuration, or the test interface permissions are not fully open.
How to confirm the configuration is complete
- Import all associated documents for a single batch of medical device production, initiate a query of "verify the supplier materials, sterilization and clean records for this batch", and check whether the number of returned citations matches the configured single search limit.
- View the
responseDatafield in the returned results to confirm that core traceability information such as batch ID and document source path is included. - Adjust the value of
quoteMaxTokento verify whether the length of returned document fragments meets the viewing and verification needs of audit personnel. - Simulate a scenario where a core document is missing, check whether the system prompts for the missing traceability source, and confirm that the prompt logic for abnormal scenarios works properly.
The values provided are common starting points and should be measured against the reader's own samples.