This page is written based on FastGPT v4.17.0, tailored for the inspection and process record audit scenario of biologics and aseptic manufacturing, and outlines the configuration, constraints, and implementation considerations for knowledge base retrieval and recall.
What the data for this category looks like
Core data for this category includes cell bank and seed lot records, culture and purification records, aseptic filling records, environmental monitoring data, media simulation filling results, aseptic test reports, and batch release documents. Data sources include cell bank and seed lot systems, process analysis equipment, clean zone monitoring systems, MES, LIMS, and equipment integrity testing tools. Update cadence falls into three categories: batch records generated per batch, clean zone monitoring and process parameters uploaded in real time, and equipment integrity data updated per fixed monitoring plans. Documents include structured fields such as batch number, culture temperature (℃), filling time, sedimentation bacteria concentration (cfu/㎡), as well as unstructured process description text.
Constraints imposed by this category's data characteristics on knowledge base retrieval and recall
The data characteristics of this category impose multiple constraints on the retrieval and recall process. Multi-source heterogeneous data sources require the retrieval system to support both structured field matching and unstructured text semantic retrieval to cover different types of audit needs. The mixed update cadence of real-time and batch-based updates requires the system to combine incremental indexing and full indexing, to ensure newly generated batch records and real-time monitoring data can be retrieved quickly. The high-risk nature of documents requires recall results to have high precision, to avoid irrelevant data interfering with the audit process. At the same time, the isolation requirement for batch records requires the retrieval scope to be limited to the dataset for the corresponding production batch, to prevent cross-batch data confusion.
How to set configurations
The following configurations are based on FastGPT v4.17.0, adapted to the audit scenario requirements of biologics and aseptic manufacturing.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for aseptic manufacturing process parameters, to avoid mistakenly recalling irrelevant batch records |
limit | Top 8 entries | Balances context length and information completeness, adapts to the multi-dimensional data needs of batch release audits |
searchMode | mixedRecall | Supports both structured field matching and unstructured text semantic retrieval, covering different types of audit documents |
usingReRank | Enabled | Re-ranks initial recall results, to improve the recall priority of high-risk process records |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands professional terms related to audits, covering scenario expressions such as "aseptic boundary" and "media simulation filling" |
collectionIds | Dataset IDs bound to production batches | Limits retrieval scope, avoids interference from cross-batch data, and complies with batch record isolation requirements |
Three common mistakes
- Issue: When using the `bge-m3 embedding model, semantic retrieval scores are abnormally high, leading to recall of a large number of irrelevant historical batch records. Cause: The matching threshold is not adjusted for professional biologics text, or professional domain text preprocessing processes are not enabled.
- Issue: Process flow charts and clean zone layout diagrams in the knowledge base cannot be displayed normally, making it impossible to view relevant content during audits. Cause: Image resource upload format and external link hosting rules are not configured correctly, leading to failure to associate image resources during retrieval.
- Issue: The number of results returned by retrieval requests is insufficient, failing to cover all process data required for audits. Cause: The
limitparameter is not set correctly, or relevant datasets are not added to thecollectionIdsscope, leading to restricted retrieval range.
How to confirm the configuration is correct
- Upload a batch of simulated cell bank records and aseptic test reports, perform a retrieval operation, and verify that the recall results include key parameters and process descriptions for the corresponding batch.
- Check the system retrieval logs, confirm that the
usingReRankfunction is active, and high-confidence process records appear at the top of the retrieval results. - Test the incremental update function, upload new media simulation filling records, and verify that the retrieval results include the newly added data content in real time.
- Check permission configurations, confirm that different audit roles can only retrieve production batch datasets bound to the corresponding
collectionIds.
The values provided here are common starting points and should be measured against the samples of the respective users.