Dermatology Workflow Orchestration

Data Characteristics

Dermatology policies and SOPs primarily originate from internal medical institution documents. These include regulations, operational guidelines, treatment protocols, and drug instructions. Documents are typically in PDF, Word, or internal HTML formats. Update frequency varies: core policies like infection control or critical value reporting may be revised annually or as needed. Drug instructions and specific treatment SOPs may change with new drug releases or technological advancements, without a fixed schedule. SOPs often feature standard sections such as "Purpose," "Scope," "Responsibilities," "Operating Procedures," and "Precautions." Procedure descriptions are detailed, containing extensive medical terminology, drug names, dosage units (e.g., mg/kg), and diagrams (e.g., skin lesion grading charts). Specific attention is required for fields and units like drug dosages, treatment durations (e.g., days, weeks), and test result indicators (e.g., U/L, ng/mL).

Constraints Imposed by Data Characteristics on Workflow Orchestration

Dermatology data characteristics impose specific requirements on workflow orchestration. First, diverse document sources and irregular update frequencies necessitate a flexible document ingestion and version control mechanism to ensure policy timeliness. Second, the structured nature of SOPs, particularly "Operating Procedures," requires fine-grained knowledge base chunking to prevent step fragmentation and maintain question-answering coherence. The abundance of specialized terminology and drug names demands accurate recognition and understanding during question-answering to avoid semantic deviations. The presence of diagrams constrains text extraction, potentially requiring image recognition or manual annotation. Furthermore, numerical information with units, such as dosages and durations, requires unit conversion and range validation during extraction and comparison to ensure accuracy. For example, 20mg and 0.02g should be recognized as equivalent.

Configuration Guidelines

Configuration Item	Recommended Value	Rationale
`Chunk size` (Chunk Length)	`800–1200 characters`	Dermatology SOPs have detailed procedure descriptions; this length ensures contextual completeness. Shorter chunks risk fragmentation, while longer ones dilute key information.
`Recall count` (Retrieval Count)	`Top 5–8 entries`	Ensures coverage of different sections from multiple related policies or SOPs, improving retrieval comprehensiveness.
`Similarity threshold` (Similarity Threshold)	`0.75–0.85`	Balances retrieval accuracy and coverage, reducing interference from irrelevant or low-relevance content.
`Rerank result count` (Reranked Return Count)	`Top 3 entries`	Focuses on the most core and accurate answer content, reducing model processing load.
`PARSE_FILE_TIMEOUT_SECONDS`	`600 seconds`	Accommodates parsing time for large PDF or Word documents, preventing file processing failures due to timeouts.
`maxContext`	`4000 tokens`	Supports questions involving multiple policies or complex SOPs, providing sufficient contextual support.

Common Pitfalls

Code modules cannot access localStorage or application URLs during workflow execution. This causes steps that retrieve shared application addresses to fail. The workflow's code execution environment is isolated from the browser environment and cannot directly access client-specific global objects.
The model generates incorrect units or numerical values when answering questions about drug dosages or treatment durations. This occurs because the original documents contain various expressions (e.g., g and mg), and knowledge base chunking fails to standardize them, leading to ambiguity during model inference.
For policy documents containing numerous diagrams, question-answering results lack visual information, leading to a poor user experience. File parsing primarily extracts text content and does not perform structured recognition or description of images, resulting in the loss of diagrammatic information.

Verification Steps

Upload representative dermatology policy documents. Review knowledge base chunking results. Verify that critical steps, drug names, and dosage units are segmented completely and accurately.
Conduct multi-round question-answering tests for dermatology policy questions of varying complexity. Evaluate the accuracy, completeness, and relevance of answers. Cross-reference with original documents for validation.
Test questions involving specific medical terminology and numerical units. Confirm the model correctly identifies, understands, and generates answers with correct units. For example, inquire about the usage and dosage of Fluticasone Propionate Cream.
Check workflow execution logs. Confirm all modules (e.g., file parsing, knowledge retrieval, model inference) execute without errors and response times are within an acceptable range.

Note: The values provided are common starting points. Measure against your own samples for optimal performance.

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.