Data Characteristics in This Category
Data in the rational drug use domain primarily originates from drug guidelines, package inserts, clinical pathways, and adverse drug reaction reports published by national drug administrations and provincial health commissions. This data updates frequently. Package inserts and adverse drug reaction reports, for example, may update monthly or even weekly. Document structures vary. They include PDF guidelines, Word-format clinical pathways, and structured database entries for drug properties and interactions. Fields cover drug generic names, brand names, indications, contraindications, dosage and administration, adverse reactions, and special population considerations. Units involve dosage (mg, g), concentration (%), and time (hours, days), requiring precise accuracy.
Constraints Imposed by These Characteristics on "Workflow Orchestration"
High update frequency demands flexible data source connectivity and scheduled synchronization capabilities in workflows to ensure real-time knowledge base content. Diverse document structures require workflows to process different file types and perform effective text extraction and structured conversion. For example, PDF drug guidelines need precise text extraction and table recognition. Word-format clinical pathways may require recognition of specific chapter headings and key information. Field and unit precision constrain the accuracy of RAG retrieval and generation. Workflows must differentiate between similar but distinct drug names or dosage units during knowledge recall. They must also avoid unit confusion or numerical errors during answer generation. Furthermore, specific drug use scenarios may involve complex logical judgments, such as drug interaction assessment. This requires workflows to orchestrate multi-step tool calls and conditional branching.
Configuration Settings
| Configuration Item | Suggested Value | Rationale |
|---|---|---|
Data Source Update Frequency | Every days 02:00 | Ensures daily retrieval of the latest drug guidelines and adverse reaction data. Avoids impacting system performance during peak daytime hours. |
Document Chunk Size | 500 characters | Balances semantic completeness and retrieval efficiency. Excessive length may introduce irrelevant information. Insufficient length may fragment critical context. |
Recall count | Top 8 entries | Considers both retrieval quality and model input length limits. Ensures coverage of highly relevant knowledge snippets. |
Similarity threshold | 0.75 | Prevents low-relevance documents from interfering with judgment. Ensures recalled knowledge highly matches the query. |
Tool Call Timeout | 60 seconds | Provides sufficient time for external tools (e.g., drug database queries) to respond. Prevents task failures due to network fluctuations. |
RAG_Model Context | 16k tokens | Accommodates potentially long context requirements in complex drug use scenarios, such as multi-drug combination evaluation. |
Common Pitfalls
- Workflow template import results in empty functional nodes. This occurs when the imported file format is incompatible or the version does not match, leading to parsing failure.
- Specific models perform poorly during tool calls, manifesting as inaccurate answers or missing key information. This often results from insufficient model fine-tuning for biomedical terminology and logic, or incorrect API parameters in tool configurations.
- The desired model cannot be selected during text content extraction. This indicates the model service is not correctly registered with the FastGPT platform, or current workflow permissions restrict model access.
Verification Steps
- Upload the latest drug package insert PDF file. Observe whether the workflow correctly extracts key fields like drug name, indications, and dosage. Cross-reference the extracted results with the original text.
- Construct a query involving complex drug interactions. Check if the workflow accurately identifies and issues warnings via tool calls. Compare the results with professional pharmaceutical knowledge.
- After a scheduled task triggers, check the data source logs. Confirm if new drug data successfully synchronized and updated the knowledge base. Randomly sample records to verify their timeliness.
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.