Data Characteristics in This Category
Vendor audit data in pharmacovigilance primarily comes from audit reports, corrective action plans, on-site inspection records, and regulatory compliance documents. These documents are often in PDF, Word, or scanned image formats, with low structural consistency. Data update frequency varies; updates typically occur after an audit cycle or when corrective actions are implemented. Audit reports may include assessments of a vendor's quality management system, production processes, and adverse event reporting mechanisms. Fields within these reports include defect types, severity, recommended corrective actions, completion dates, and responsible parties. Most units are text descriptions. Dates are in date format, and quantities are integers or percentages.
Constraints on "Deployment and Upgrade" from These Characteristics
The unstructured nature of audit reports demands advanced document parsing capabilities from FastGPT. Deployment requires configuring an OCR plugin to process scanned documents and optimizing the segmentation strategy to prevent key information from being fragmented. The uncertain update frequency means that model training and knowledge base synchronization should not be too frequent. An event-driven or periodic incremental update model is appropriate to avoid resource waste. Documents contain extensive specialized terminology and regulatory clauses, requiring the base model to have strong domain understanding. Terminology glossaries should be imported during knowledge base construction. The diversity of field descriptions, such as qualitative descriptions of defect severity, affects recall and matching accuracy. This requires optimizing through similarity thresholds and re-ranking strategies.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
UPLOAD_FILE_MAX_SIZE | 200 MB | A single audit report file can be large, containing multiple attachments. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Accommodates parsing time for complex or scan-heavy documents, preventing timeouts. |
Chunk size | 800–1200 characters | Maintains semantic completeness, balancing long descriptions and detailed information. |
Recall count | Top 10 entries | Ensures coverage of potentially dispersed key information in vendor audit reports. |
Similarity threshold | 0.75 | Addresses specialized terminology and descriptive diversity, improving matching accuracy. |
Rerank result count | Top 3 entries | Focuses on the most relevant audit findings or corrective actions, reducing irrelevant interference. |
Three Common Mistakes
- After a knowledge base update, query results still show old information. The knowledge base synchronization strategy was not configured for incremental updates, leading to a mix of old and new data.
- An "file too large" error appears when uploading an audit report file. The
UPLOAD_FILE_MAX_SIZEparameter is set too low and does not accommodate the actual file size. - The chat interface fails to load on iOS devices, displaying a white screen. Front-end resource loading paths are misconfigured, or cross-origin policies restrict mobile access.
How to Verify Correct Configuration
- Upload an audit report PDF file containing scanned images. Confirm that the file parses correctly and generates Q&A content.
- Query for audit findings regarding a specific vendor. Verify that the returned results accurately include defect numbers and corrective actions from the report.
- Simulate uploading a new audit report. Confirm that the knowledge base content is updated and that new and old information is correctly differentiated.
The values provided are common starting points and should be measured against specific 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.