Workflow Orchestration for Intelligent Triage Quality

Data Characteristics

Intelligent triage systems process data from internal medical guidelines, disease diagnosis and treatment pathways, drug inserts, medical handbooks, and historical medical records. These documents are typically PDFs, Word files, or structured database entries. Data updates are relatively stable, occurring periodically (quarterly or annually) when new drugs are approved or treatment guidelines are revised. Document structures include disease overviews, diagnostic criteria, differential diagnoses, and treatment plans in guidelines, and indications, dosage, and side effects in drug inserts. Fields involve disease codes (e.g., ICD-10), generic drug names, dosage units (mg, ml), and treatment durations (days, weeks). Accuracy and consistency requirements are high.

Constraints on Workflow Orchestration

The data characteristics of intelligent triage quality documentation impose several constraints on workflow orchestration. The authoritative nature of diagnostic guidelines requires precise knowledge base recall. Ambiguous or misleading information is unacceptable. This mandates strict similarity thresholds for retrieval and recall nodes in the workflow. Document update frequency is not high, but each update can involve substantial content. The workflow needs to support batch document processing and version management to ensure real-time accuracy of knowledge base content. The specialized nature of fields and the strictness of units, such as drug dosage and treatment duration, require the workflow to accurately identify and maintain data integrity during information extraction and structuring. This prevents triage errors due to unit conversion or missing fields. Additionally, for compliance, the workflow must trace the knowledge source for each triage recommendation and support manual review and intervention.

Configuration Guidelines

Configuration Item	Suggested Value	Rationale
`Chunk size` (Segment Length)	`300-500 characters`	Diagnostic guidelines usually have clear paragraph structures. Segments that are too long introduce irrelevant information, while segments that are too short may lose context and semantic completeness.
`Recall count` (Recall Count)	`Top 5-8 entries`	This ensures coverage of core diagnostic recommendations or drug information while avoiding excessive redundancy and improving model processing efficiency.
`Similarity threshold` (Similarity Threshold)	`0.75-0.85`	Intelligent triage demands high information accuracy. A higher threshold filters out irrelevant recall results, reducing the risk of misinformation.
`Rerank result count` (Rerank Return Count)	`Top 3 entries`	This further refines recall results, presenting the most relevant information to the large language model and improving the accuracy of final triage recommendations.
`Max Concurrent Documents`	Calibrated by actual measurement	To handle quarterly or annual batch document updates, balance system resources with processing efficiency to ensure a smooth update process. For example, the `max_concurrent_docs` configuration item.
`External Function Call Timeout`	`60 seconds`	This allows sufficient response time for potential calls to external drug databases or disease diagnosis interfaces, preventing workflow interruptions due to external service delays. Corresponds to the `function_call_timeout` parameter.

Common Pitfalls

During workflow execution, triage recommendations may have low relevance to user symptoms. This occurs when the Similarity threshold (Similarity Threshold) is set too low, leading to the recall of many irrelevant document segments.
Errors or omissions in drug dosage units appear in triage recommendations. This happens when the document parsing node fails to correctly identify or extract specific fields from drug inserts, such as an empty dosage_unit field.
The workflow encounters numerous parsing failures when processing new versions of diagnostic guidelines. This indicates that the document's internal structure or formatting changed after the version update, rendering the existing Document Parsing Rules ineffective.

Validation Steps

Select representative diseases and symptoms. Conduct multiple rounds of intelligent triage testing. Verify that triage results align with official diagnostic guidelines.
Randomly select a batch of drugs. Simulate user inquiries. Check the accuracy of drug names, dosages, and usage instructions in the triage recommendations.
After document updates, execute a complete knowledge base reconstruction workflow. Check logs for warnings about file parsing failures or missing content, such as the parsing_error_count field.

Note: The values provided are common starting points. Measure performance against specific samples to determine optimal configurations.

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.