Data Characteristics in This Category
Pharmacovigilance data in health management primarily originates from electronic health records (EHRs), wearable device data, medication usage records, and patient-reported adverse event reports. This data typically combines structured and unstructured formats. EHRs contain structured data like physician orders, diagnoses, and medication records. Free-text descriptions of conditions, consultation notes, and adverse reaction symptoms are unstructured. Data updates frequently; EHRs update in real-time with clinical activities, and wearable device data may upload minute-by-minute. Document formats are diverse, including PDF drug inserts, clinical guidelines, and medical records in HL7 CDA or FHIR formats. Fields and units are medically specific, such as dosage units like mg/kg, time units like hours/day, and disease/symptom coding using ICD-10 or SNOMED CT.
Constraints Imposed by These Characteristics on "Tool Calling and Plugins"
The high update frequency and diversity of health management pharmacovigilance data require tool calling and plugins to possess real-time data processing capabilities. This ensures rapid response to the latest adverse reaction information. The coexistence of structured and unstructured data means plugins must parse standardized medical codes and extract key information from free text using natural language processing (NLP) techniques. For example, identifying adverse reaction symptoms like "dizziness" or "nausea" from patient self-reports. Medically specific fields and units demand high accuracy for tool calling parameter validation and result interpretation. This requires accurate unit conversion to avoid misjudgments due to unit confusion. The variety of document formats necessitates plugins with multi-format file parsing capabilities, such as extracting drug contraindication information from PDF inserts. These constraints collectively dictate that configuring tool calling in FastGPT requires a refined design for data sources, data formats, and data processing logic.
Configuration Settings
| Configuration Item | Suggested Value | Rationale |
|---|---|---|
maxContext | 8000 tokens | Balances the need for long medical records and real-time interaction, preventing context truncation. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Processing large PDF drug inserts or complex medical reports can be time-consuming. |
Recall Count | 10 items | Ensures retrieval of sufficient relevant pharmacovigilance information from the knowledge base, improving accuracy. |
Similarity Threshold | 0.75 | Medical terminology requires high precision matching, reducing interference from irrelevant information. |
tool_request_timeout | 60 seconds | Most external medical service interfaces respond within this timeframe, preventing premature timeouts. |
function_call_model | gpt-4o | Offers stronger function call understanding for complex medical scenarios, increasing plugin call success rate. |
Three Common Mistakes
- External service calls return 400 or 500 error codes: This usually indicates that the input parameter format does not conform to the external API interface requirements, such as incorrect medical codes (e.g.,
ICD-10) or incorrect date formats. - Model output results lack critical information or are inaccurate: This often occurs when knowledge base segment length is inappropriate, leading to truncation of relevant context and failure to provide complete information for the model to infer.
- Plugin calls result in long delays or timeouts: Possible reasons include slow responses from external service interfaces or a
tool_request_timeoutparameter set too short in FastGPT.
How to Confirm Proper Configuration
- Use FastGPT's debugging interface to verify that tool call input parameters exactly match the format defined in the external API documentation.
- Manually construct medical data for typical patient cases and simulate conversations. Check if the model correctly identifies adverse reactions and calls relevant tools to obtain drug contraindications or treatment suggestions.
- Check log output to confirm the status codes and response bodies returned by external services, ensuring correct data parsing.
- For specific drug adverse reaction queries, compare FastGPT's results with official drug inserts or professional database information to assess accuracy.
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.