Data Characteristics for this Category
Deviation and Corrective and Preventive Action (CAPA) registration and declaration documents typically include unstructured content. Examples are deviation investigation reports, CAPA plans, implementation records, and effectiveness verification reports. These documents are primarily in PDF or Word format. Their internal structure includes titles, body text, attachments, and signature pages. Data sources are diverse, including quality management systems, manufacturing execution systems, and laboratory information management systems. The update frequency is relatively low, usually occurring after a deviation event or at the end of a CAPA cycle. Documents contain extensive specialized terminology, abbreviations, dates, batch numbers, and equipment numbers. Units may involve dosage units (e.g., mg), time units (e.g., hours), and concentration units (e.g., %w/v). Accurate identification of these elements is critical.
Constraints Imposed by these Characteristics on "Deployment and Upgrade"
The unstructured nature of deviation and CAPA documents demands advanced document parsing capabilities during FastGPT deployment, especially for recognizing text within tables and images. The low update frequency means a large initial data import for knowledge base construction, but less pressure for incremental updates later. Deployment should focus on the efficiency and stability of initial data import. The prevalence of specialized terminology and abbreviations in documents can lead to misunderstandings by general language models. Fine-tuning or enhanced retrieval is necessary to improve semantic understanding. Accurate identification of key fields and units is fundamental for compliance in registration and declaration documents. Model training and knowledge base configuration must ensure precise extraction and matching of this information. Given data sensitivity, private deployment is preferred, requiring attention to system integration and security configurations.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
UPLOAD_FILE_MAX_SIZE | 500 MB | Deviation and CAPA reports may contain numerous images and attachments, leading to large individual files. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing large PDF or Word documents can be time-consuming. This prevents timeouts that lead to parsing failures. |
Chunk size (Segment Length) | 800–1200 characters (characters) | Ensures each text segment contains sufficient contextual information while avoiding excessive length that impacts retrieval efficiency. |
Recall count (Recall Count) | Top 10 entries (top 10) | Increases the coverage of retrieval results, ensuring highly relevant document snippets are recalled. |
Similarity threshold (Similarity Threshold) | Calibrate based on actual measurements | Adjust based on actual corpus and business needs to ensure precision of recall results. Start testing from 0.75. |
MAX_MEMORY_SIZE | 16 GB | Ensures sufficient memory when processing complex queries and large knowledge bases, preventing system crashes. |
Common Pitfalls
- Knowledge base query response times are excessively long or result in timeout errors. This can occur in private deployments if
MAX_MEMORY_SIZEis configured too low, leading to insufficient memory when processing large volumes of documents and complex queries. - Model responses show misunderstandings or confusion regarding specialized terminology. This happens when there is insufficient data augmentation or fine-tuning for the specific terminology of the deviation and CAPA domain. General models then fail to accurately comprehend their meaning.
- Key information (e.g., batch numbers, dates) is incomplete or incorrectly formatted after document parsing. This indicates that the document parsing component configuration does not adequately recognize all report formats. Examples include insufficient OCR recognition rates for scanned documents or imprecise regular expression matching.
How to Confirm Proper Configuration
- Upload typical large deviation investigation reports and CAPA plan documents. Check the completeness and readability of parsed text segments.
- Use query statements containing specific specialized terminology and abbreviations. Verify the accuracy of model responses and cross-reference them with original documents.
- In the application debugging interface, enter questions containing key fields (e.g., batch numbers, dates, equipment numbers). Check if the model can correctly retrieve and cite this information from the knowledge base. Set an acceptable error rate threshold that meets business requirements.
Note: 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.