Multi-Turn Conversations and Prompts for Respiratory Quality Documentation

Data Characteristics

Quality documentation related to respiratory diseases draws from various sources. These include clinical guidelines, drug inserts, medical device registration certificates, adverse event reports, Standard Operating Procedures (SOPs), and regulatory documents. Update frequencies vary; clinical guidelines and drug inserts may update annually or biennially, while adverse event reports generate in real-time. Document structures typically include text, tables, charts, and appendices. Fields and units are highly specialized, for example, dosage (milligrams, micrograms), lung function indicators (liters, milliliters/second), diagnostic criteria (kPa, mmHg), and specific medical terms and abbreviations. Key information often scatters across different sections and involves extensive cross-references.

Constraints on Multi-Turn Conversations and Prompts

The specialized and diverse nature of respiratory quality documents demands high contextual understanding in multi-turn conversations. Numerous medical terms and abbreviations require accurate terminology parsing to prevent semantic drift. Inconsistent document update frequencies necessitate a version management mechanism for the knowledge base, ensuring conversations rely on the latest valid information. The scattered distribution of key information in complex document structures requires prompt design to guide the model in deep semantic association and multi-source information integration. For example, when a user asks about drug contraindications for a specific respiratory disease, the model must recall information from drug inserts and may also need to reference relevant clinical guidelines, comparing and synthesizing information from different documents. The strictness of fields and units requires the model to accurately cite or convert them in responses, avoiding misunderstandings due to unit confusion.

Configuration Recommendations

Common Pitfalls

• Symptom: Conversation interface returns numerous irrelevant auxiliary details or source tracing symbols. Reason: Prompts fail to effectively constrain the model's output format, or debugging information is not disabled in the model configuration.
• Symptom: When users ask about diagnostic criteria for a respiratory disease, the model's response lacks critical numerical values or units. Reason: The knowledge base construction did not structure tables or key fields, preventing the model from accurately extracting numerical values with units.
• Symptom: In multi-turn conversations, the model fails to correctly understand a user's follow-up question about drug dosage mentioned in a previous turn. Reason: The maxContext parameter is set too low, causing the model to lose critical historical context in multi-turn conversations.

How to Validate Configuration

• Design a series of multi-turn conversation test cases for typical respiratory diseases (e.g., asthma, COPD) diagnosis, treatment, and medication scenarios. Verify if the model accurately understands and provides complete responses.
• Select document snippets containing complex tables and charts. Test if the model can extract key numerical values and units and apply them correctly in conversation responses.
• Simulate users repeatedly mentioning specific medical terms or abbreviations during a conversation. Check if the model consistently maintains correct understanding and consistent responses for these terms.

Note: The values provided are common starting points. Measure them 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.