What This Category’s Data Looks Like
Typical documents for this category include cell bank or seed lot records, culture and purification records, aseptic filling records, environmental monitoring data, media fill simulation results, sterility test reports, and batch release documents. Data sources cover cell bank and seed lot systems, process analysis equipment, clean zone monitoring systems, MES, LIMS, and equipment integrity testing tools. Data update frequencies fall into two groups: updates per production batch, real-time monitoring updates, and updates per fixed monitoring plans. Documents contain structured fields and unstructured descriptions. Structured fields such as culture temperature and settleable plate count include professional units. Unstructured content includes process steps and exception handling workflows.
Constraints Imposed on Knowledge Base Retrieval and Recall
Dispersed data sources and varied update frequencies for this category require retrieval and recall functions to support cross-source data integration and incremental updates. This prevents missed real-time clean zone monitoring data or batch release documents updated per batch. Documents include structured parameters and unstructured processes. Retrieval functions must match both exact fields and semantic associations. This ensures accuracy of process parameters and completeness of process descriptions. The specificity of professional terms and units requires retrieval models to adapt to biologics domain vocabulary. This avoids invalid recall results from term mismatches. Incorrect retrieval results may cause process document deviations. This can lead to aseptic assurance failure. Strict control of recall result relevance is required.
How to Set Configurations
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches precise associations between biologics professional terms and process parameters, avoids low-relevance results interfering with process judgment |
limit | Top 8 results | Individual process documents have dense content. Too many recall results will exceed the context window and reduce drafting efficiency |
searchMode | mixedRecall | Use mixed semantic/vector and full-text retrieval. Exact-field, metadata, batch, and document-type constraints require separate filters and are not enforced by this mode. |
usingReRank | Enabled | Filters redundant or low-relevance documents from cross-source retrieval, improves professionalism of recall results |
datasetSearchUsingExtensionQuery | Enabled | Expands professional terms to match exclusive record fields for specialized scenarios including cell banks and aseptic filling |
collectionIds | Bind exclusive document collections for vaccines/aseptic injectables | Only recall data exclusive to this category, avoids mixing irrelevant documents that interfere with drafting |
Three Common Configuration Mistakes
- Phenomenon: Knowledge base content is not synced to the deployed database. The interface displays blank or missing documents. Cause: Database connection parameters are not configured correctly, or the data synchronization script is not executed.
- Phenomenon: Custom-split document chunks are automatically deduplicated. This causes the original document’s index order to not match the split results. Cause: The knowledge base enables global deduplication by default, and the configuration is not adjusted for this category’s batch records.
- Phenomenon: Retrieval results do not return expected process parameter records. Returns empty results or irrelevant content. Cause: Exclusive document collections are not bound, or the similarity threshold is set too high, filtering valid matching results.
How to Verify Correct Configuration
- Run a test retrieval, enter professional terms related to biologics processes. Verify that returned results come from the bound exclusive document collection.
- Check the number of retrieval results. Confirm it falls within the configured single search upper limit range.
- Verify that the re-ranking function is active. Compare result differences between enabling and disabling re-ranking to confirm alignment with business requirements.
- View database synchronization logs. Confirm that custom-split document chunks have been imported correctly, with no order anomalies caused by automatic deduplication.
The values provided are common starting points and should be measured against the reader's own samples.