What Does Data for This Category Look Like
The data for this category is sourced from approved formulation processes, material batches, MES/electronic batch records, equipment and environmental monitoring systems, LIMS, and quality management systems. Daily documents include master production files, batch production records, cleaning records, inspection records, deviation, CAPA, and change documents, and batch release approval forms, which are updated for each batch production, each key process step, and in accordance with monitoring plans. Document fields include material batch numbers, process parameter values, test results, deviation descriptions, and other items, with corresponding units covering production-related metrological standards such as mass, temperature, and duration.
What Constraints Do These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
The batch-by-batch and step-by-step updated data characteristics of this category require that retrieval must be tied to the material batch number and process parameters of the current production batch, to avoid calling expired or mismatched historical data. Structural differences across multiple types of compliance documents require that retrieval supports filtering target fields by document category, to ensure matched content aligns with the current review stage such as material feeding review and deviation summary. The high-frequency update characteristic requires that the knowledge base synchronization frequency matches the production rhythm, while the retrieval scope must be limited to only documents that have completed quality release to comply with GMP requirements. If retrieval matches incorrectly, it will directly affect batch record review and release decisions, leading to compliance risks.
How to Set Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Adapts to the field precision requirements of pharmaceutical dosage form manufacturing documents, filters irrelevant historical data with low matching degrees |
limit | Top 6–10 results | Covers multi-dimensional review requirements for current batch materials, processes, deviations, and other aspects, avoids excessive results increasing review workload |
searchMode | mixedRecall | Balances semantic matching and precise field matching, adapts to retrieval requirements for multiple types of compliance documents |
usingReRank | Enabled | Sorts initial retrieval results by relevance, ensures documents most relevant to the current review stage are displayed first |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands keywords related to reviews, covers associated documents such as deviation handling and batch release |
collectionIds | Bind to the knowledge base collection corresponding to the current production batch | Limits retrieval scope to only compliance documents of the current project, avoids interference from cross-batch data |
Three Common Misconfigurations
- API calls that nest knowledge base assistant workflows return empty values. Cause: FastGPT v4.17.0 or a later version is not being used, or the
collectionIdsparameter is not correctly configured to bind the knowledge base collection corresponding to the current production batch in the workflow. - A quote type error occurs when referencing knowledge base variables. Cause: The incoming retrieval variable format does not meet system requirements, does not follow the preset parameter structure, or the variable contains unprocessed special characters.
- Retrieval results do not cover deviation or release-related documents for the current batch. Cause: The
datasetSearchUsingExtensionQueryparameter is not enabled, so associated keywords cannot be automatically expanded, resulting in missed review content.
How to Verify Proper Configuration
- View the knowledge base retrieval node in the workflow, confirm that the
collectionIdsparameter is bound to the knowledge base collection corresponding to the current production batch. - Perform a simulated retrieval, check whether the matching degree of returned results meets the precision requirements of the current review stage, and adjust the
similaritythreshold to an appropriate range. - Trigger an API call workflow, confirm that the returned results are not empty and contain relevant document content for the current batch.
- Check the sorting of retrieval results, confirm that the
usingReRankparameter is active, with the most relevant documents displayed first.
The values provided here are common starting points and should be evaluated against the reader's own samples.