Data Characteristics in this Category
Infection control data originates from Hospital Information Systems (HIS), Laboratory Information Systems (LIS), Electronic Medical Records (EMR), and specialized infection surveillance systems. This data is highly real-time and distributed. Examples include microbial culture results, antibiotic susceptibility test reports, infection site diagnoses, and patient length of stay. Data updates frequently; some monitoring data may update hourly. Document structures vary, including standardized lab reports, unstructured physician progress notes, and structured case admission information. Fields and units involve bacterial names, antibiotic types, MIC (Minimum Inhibitory Concentration, unit μg/mL), infection rates (percentage), and incidence rates (per thousand patient-days). The data contains extensive medical terminology and abbreviations, requiring high accuracy in data parsing.
Constraints on Tool Calling and Plugins
The real-time nature of infection control data requires tool calling plugins to respond quickly, supporting immediate alerts and interventions. Distributed data sources necessitate integrating multiple data interfaces; tool calling must flexibly adapt to different system API specifications and authentication methods. Diverse document structures challenge data extraction and parsing. Unstructured text requires advanced Natural Language Processing (NLP) capabilities for structured conversion, such as identifying infection sites and pathogens from progress notes. The presence of specialized fields and units means tools must correctly identify and apply medical unit conversion rules when processing numerical data to avoid misinterpretations. Furthermore, extensive medical terminology and abbreviations require the large language model to accurately understand context when calling external tools for queries or calculations, preventing incorrect calls or result deviations due to semantic ambiguity.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 8000–12000 tokens | Accommodates long text descriptions in infection control medical records, ensuring context completeness |
API_TIMEOUT_SECONDS | 60 seconds | Ensures timely returns for real-time infection control data queries, preventing system blockage |
toolCallRetries | 3 times | Handles occasional network fluctuations or transient unavailability of external services, improving call success rate |
knowledgeBaseThreshold | 0.75–0.85 | Accurately matches professional terms and standards in the infection control knowledge base, reducing false recalls |
toolFunction | Specific API interface, e.g., queryInfectionData | Directly interfaces with infection surveillance systems to obtain real-time infection data and pathogen information |
parserConfig | JSON schema, including pathogen, antibiotic, MIC | Structurally parses laboratory reports, extracting key fields such as pathogen, antibiotic, and minimum inhibitory concentration |
Three Common Mistakes
- Symptom: Tool call returns
500 Internal Server ErrororConnection Timeout. Reason: The external infection surveillance system API is slow or the network is unstable, and insufficient timeout or retry mechanisms are configured. - Symptom: The infection rate calculation result output by the model is inconsistent with the actual situation, or the recommended antibiotics are inaccurate. Reason: The tool fails to correctly identify units (e.g.,
mg/dLvs.g/L) or map abbreviations (e.g.,MRSA) to their correct full names when parsing numerical fields in medical reports. - Symptom: In advanced orchestration, a knowledge base search plugin is configured, but the model does not reference knowledge base content when answering related questions. Reason: The knowledge base recall threshold is set too high, causing valuable documents with low similarity to the user's query to not be recalled.
Verification of Configuration
- Simulate actual infection control consultation scenarios to check if tool calls accurately retrieve patient infection data, microbial culture results, and antibiotic susceptibility test reports. Verify the completeness and accuracy of the returned data.
- Cross-reference medical terminology, units, and numerical values in reports or recommendations generated by the model after tool calls with the original data. Pay special attention to key indicators such as
MICvalues and infection rates. - In FastGPT's logs or debugging interface, inspect the
request bodyandresponseof each tool call to confirm correct parameter passing and that the external service's returned data format is as expected.
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.