Data Characteristics for this Category
Deviation and Corrective and Preventive Action (CAPA) data in the biopharmaceutical sector typically originates from exported files from Quality Management Systems (QMS), Manufacturing Execution Systems (MES), or Laboratory Information Management Systems (LIMS). This data updates infrequently, usually archived by batch, event, or monthly. Document structures are primarily structured tables or semi-structured reports. They include fields such as event descriptions, root cause analyses, corrective actions, preventive actions, responsible parties, and completion dates. Key fields like Deviation Level and Root Cause Classification often use enumerated values. Impact Assessment may contain free-text descriptions. Time fields such as Occurrence Date and Closure Date maintain a consistent format. However, the Impact Description field varies significantly in length and complexity.
Constraints Imposed by these Characteristics on "Deployment and Upgrade"
The structured nature of Deviation and CAPA data sources favors a table or structured document parsing strategy for knowledge base construction. This preserves the relationships between fields. The relatively low update frequency means the knowledge base synchronization mechanism can be set for periodic full updates or incremental updates, eliminating the need for real-time synchronization. Free-text descriptions within documents, such as Impact Assessment and Root Cause Analysis, require specific segmentation strategies. These strategies must balance contextual completeness with retrieval efficiency. The presence of enumerated value fields allows for preliminary filtering using exact matches or filters during retrieval, improving recall accuracy. Additionally, the data may contain specialized terminology and abbreviations. This necessitates pre-building a domain dictionary or standardizing terminology to optimize vector embedding performance.
Configuration Settings
| Configuration Item | Recommended Value | Basis for Recommendation |
|---|---|---|
Segment Length | 500–800 characters | Balances contextual completeness of long text descriptions with retrieval granularity |
Overlap Length | 50–100 characters | Ensures contextual continuity at segment boundaries, reducing information loss |
Recall Count | Top 8–12 items | Provides sufficient initial relevant information, considering query complexity |
Similarity Threshold | 0.75–0.82 | Balances recall rate and precision, reducing irrelevant results |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Addresses potential parsing time for large report files |
UPLOAD_FILE_MAX_SIZE | 100 MB | Allows uploading large batch data files containing historical records |
Three Common Mistakes
- Query results display old information after knowledge base content updates. This occurs because the knowledge base index was not rebuilt in time or the cache was not refreshed.
- Queries for "root cause" return many irrelevant results. This usually happens when the segmentation strategy is too coarse, causing key fields to mix with irrelevant text.
- In a private deployment environment, connection to external large model APIs fails. This is due to server network configuration restrictions on external access or incorrect proxy settings.
How to Confirm Correct Configuration
- Upload Deviation and CAPA reports of different structures and sizes. Check if files parse and segment successfully. Observe if segment content is reasonable.
- For typical CAPA queries (e.g., "root cause of deviation for a specific product batch"), verify if returned knowledge snippets are accurate and contain key information.
- In an intranet environment, use
pingorcurlcommands to test network connectivity between the FastGPT deployment server and the large model API service. Ensure API key configuration is correct.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-21.