Workflow Orchestration for Hematologic Oncology Quality

Data Characteristics in This Category

Quality documents in hematologic oncology primarily source data from clinical trial protocols, investigator brochures, drug labels, Standard Operating Procedures (SOPs), and regulatory documents like GMP and GCP guidelines. These documents have a relatively low update frequency, typically undergoing annual or quarterly adjustments due to drug development progress, regulatory revisions, or clinical practice updates. Document structures are highly standardized, often using chapter numbering, section headings, and lists. They contain highly specialized fields. For example, "dosage unit" is often mg/kg or mg/m², "adverse event grade" frequently uses CTCAE (Common Terminology Criteria for Adverse Events) grades, and biomarker detection results include BCR-ABL fusion gene or FLT3-ITD mutation. Numerical data often includes clear units of measurement and reference ranges.

Constraints Imposed by These Characteristics on Workflow Orchestration

The standardized structure and specialized fields of hematologic oncology quality documents require high-precision positioning capabilities from text extraction components within the workflow. This capability identifies key information in specific chapters or paragraphs, such as extracting drug administration regimens or adverse event reporting requirements. The low update frequency allows for deeper pre-processing and embedding during knowledge base construction. However, the workflow must identify and handle version differences to avoid referencing outdated information. Specialized fields and units demand higher accuracy from Named Entity Recognition (NER) and information extraction in the workflow. This requires configuring customized entity recognition models or dictionaries to correctly identify specific markers like BCR-ABL and associate them with corresponding test results. Additionally, the need for multi-document referencing and cross-validation requires the workflow to support complex knowledge graph construction or multi-hop reasoning. This establishes logical connections across multiple documents, for example, deriving dosage limits in clinical trial protocols from drug labels.

Configuration Guidelines

Configuration Item	Suggested Value	Rationale
`Chunk size` (Chunk Size)	`500–800 characters`	Quality documents have high content density; appropriate chunk size helps maintain contextual completeness.
`Recall count` (Recall Count)	`top 8–12 items`	Ensures coverage of key information from multiple relevant documents, especially when regulatory references are involved.
`Similarity threshold` (Similarity Threshold)	`0.75–0.85`	Guarantees high relevance of recalled results to hematologic oncology professional terminology, filtering out irrelevant segments.
`Rerank result count` (Rerank Return Count)	`top 5 items`	Focuses on the most relevant, high-quality segments, reducing noise for the large language model.
`PARSE_FILE_TIMEOUT_SECONDS`	`600 seconds`	Handles parsing of large SOPs or clinical trial protocols, preventing timeouts.
`maxContext`	`4000–8000 tokens`	Accommodates complex queries and multi-document context, meeting information association needs in hematologic oncology.

Three Common Pitfalls

Symptom: During workflow execution, the text content extraction component returns empty field values. Reason: The knowledge base's chunking strategy failed to effectively capture specific professional fields, or the Named Entity Recognition model was not trained on specific hematologic oncology terminology.
Symptom: During workflow performance testing, the maximum tps (transactions per second) is significantly lower than expected, with request queues building up. Reason: The Recall count (Recall Count) and Rerank result count (Rerank Return Count) were set too high during the knowledge base query phase, leading to a large amount of text processing for each request and increasing computational burden.
Symptom: During workflow debugging, an error message appears in version 4.8.10 stating "Text content extraction component does not support extraction from knowledge base references." Reason: This version's component parameter configuration or internal logic limitations prevent it from directly using the output of an upstream knowledge base component as input for secondary extraction. The workflow design needs adjustment, for example, by merging knowledge base reference results into a single text before extraction.

How to Confirm Proper Configuration

Select representative clinical trial protocols, SOPs, and drug labels from hematologic oncology. Verify that the workflow accurately extracts key information such as experimental drug dosage, adverse event grades, and treatment cycles.
Randomly select more than 10 queries containing specialized terminology. Check if the knowledge base segments recalled by the workflow contain the query keywords and their context. Evaluate the relevance threshold of the recalled results.
Simulate real-world usage scenarios. Conduct concurrent stress tests on the workflow. Observe the tps curve and response times. Evaluate its stability and performance under expected load. Ensure parameters like PARSE_FILE_TIMEOUT_SECONDS and maxContext do not cause bottlenecks.
For specific FastGPT platform versions, consult official documentation or community discussions. Confirm compatibility between workflow components, especially data input/output format requirements. This avoids functional limitations due to version differences.

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.