What Data Looks Like for This Category
Data for this category comes from cell banks and seed lots, process analysis systems, cleanroom monitoring data, MES, LIMS, and equipment integrity test results. Daily documents include cell bank or seed lot records, culture and purification records, aseptic filling records, environmental monitoring reports, media fill simulation records, sterility test reports, and batch release documents. Data updates occur per production batch, when triggered by real-time monitoring, or per fixed monitoring plans. Individual documents typically include fields such as batch number, process parameters, test values, and operator records. Some parameters must use statutory units: temperature in degrees Celsius, colony count in CFU per plate.
Constraints Imposed on Knowledge Base Retrieval and Recall
The multi-source, heterogeneous nature of data in this category requires retrieval to support multiple formats, including structured process records and unstructured test reports. Each batch or real-time updated data must have timely indexing to avoid recalling expired information. The strong correlation between production batches requires retrieval results to precisely match all associated documents for the target production batch, preventing cross-batch data confusion. Process parameters and test values in individual documents must strictly match statutory units and field definitions. Retrieved content must be directly usable for quality problem traceability, with no missing fields or unit errors. Additionally, the serious consequences associated with quality issues require retrieved information to have high credibility, avoiding interference from irrelevant or incorrect information during analysis.
Configuration Settings
The following configurations are based on FastGPT v4.17.0, tailored to the quality traceability and nonconformance analysis needs of this category.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | This category requires precise matching of process parameters and batch information to avoid interfering with traceability analysis via low-similarity irrelevant documents |
limit | Top 8–12 results | The number of documents associated with a single production batch is limited. Retrieving too many will increase context processing load, while retrieving too few may miss critical traceability information |
searchMode | embedding | This category requires retrieval based on semantic content such as process parameters and test values. Vector search is better suited for semantic matching needs of unstructured documents |
usingReRank | Enabled | Semantic similarity scores for multi-source heterogeneous data may have deviations. Reranking improves the ranking priority of critical information, meeting the precision requirements for quality traceability |
datasetSearchUsingExtensionQuery | Enabled | Expanding queries can optimize the retrieval scope to cover all potential documents associated with a production batch, avoiding missing traceability information across systems |
collectionIds | Bind to the category-specific knowledge base collection | Isolate documents from other categories to ensure retrieval results only include records related to the organization’s biologics and aseptic manufacturing, improving traceability efficiency |
Three Common Misconfigurations
- After importing Excel files into the knowledge base, the automatic splitting result merges multiple rows into a single entry, failing to split one row per entry. This occurs when the custom delimiter is not configured correctly, or when splitting by row is not specified as the basis, causing the system’s default chunking rules to not match the Excel row structure.
- Retrieval speed is slow for document collections split based on QA pairs, failing to meet real-time traceability requirements. This happens when the
limitandusingReRankconfigurations are not optimized for this category’s batch-associated documents, or when no dedicated knowledge base collection is bound, leading to an overly large retrieval scope. - An
upstream connect error or disconnect/reset before headerror occurs when uploading files to the knowledge base. This is caused by a connection timeout between the knowledge base service and file storage service, or large-volume process documents exceeding the default single-upload limit, leading to connection interruption.
How to Verify Proper Configuration
- During retrieval testing, a production batch’s number is input, and retrieval results are checked to confirm they only include associated documents for that batch, with no cross-batch irrelevant content.
- Process parameter keywords are input, and retrieval result ranking is checked to confirm it aligns with business priorities, with critical documents appearing at the top.
- The bound knowledge base collection configuration is checked to confirm it only includes dedicated documents for this category such as cell bank records and test reports.
- A test Excel file is uploaded, and automatic splitting results are checked to confirm they generate independent entries per row, matching the custom delimiter configuration requirements.
The values given are common starting points and should be measured against the reader's own samples.