Data Characteristics for This Category
Telemedicine product data primarily originates from online patient consultation records, smart wearable device monitoring data, electronic medical record summaries, drug instructions, and various medical knowledge bases. Data updates frequently; consultation records and monitoring data might be real-time, while drug information and medical guidelines update periodically. Document structures vary: consultation records are often unstructured text, monitoring data is structured time-series data, and drug instructions and medical record summaries fall between these, containing structured fields and free-text descriptions. Fields may include symptom descriptions, diagnostic results, medication history, allergy history, vital signs (e.g., heart rate, blood pressure, blood glucose), and examination report values. For units, vital sign data clearly specifies units (e.g., mmHg, bpm, mmol/L), and drug dosages require precision down to mg or ml.
Constraints Imposed by These Characteristics on "Multi-turn Conversations and Prompts"
The high update frequency of telemedicine data requires the knowledge base to support rapid synchronization and indexing, ensuring the timeliness of information cited in multi-turn conversations. Unstructured consultation records and semi-structured medical record summaries demand higher accuracy in text segmentation and entity extraction to identify key medical concepts. Diverse fields and units, especially vital signs and drug dosages, necessitate prompt designs that can precisely identify and perform unit conversions or verifications to avoid ambiguity. Conversations may involve patient privacy and medical risks, imposing strict constraints on the accuracy and rigor of model output. Prompts must guide the model to clarify or suggest seeking professional medical assistance when uncertain. Additionally, multi-turn conversations in telemedicine scenarios often extend over a long duration, making context management critical to maintain conversational coherence effectively.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 8 | Telemedicine conversations typically require a longer context to maintain coherence and avoid losing critical information. |
Chunk size | 400–600 characters | Balances semantic completeness of text with retrieval efficiency, preventing excessive segmentation that loses context. |
Recall count | 8–12 entries | Ensures coverage of relevant information from diverse and heterogeneous data sources, improving answer accuracy. |
Similarity threshold | 0.75–0.85 | Balances recall and precision, reducing interference from irrelevant information and minimizing hallucinations. |
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | Telemedicine documents may include large electronic medical records or examination reports, requiring relatively longer parsing times. |
Rerank result count | 5 entries | Re-filters initial retrieval results to focus on the most relevant content, enhancing the quality of the final answer. |
Common Pitfalls
- After uploading an attachment, the system does not analyze its content, preventing the attachment information from being cited in conversations. This occurs because the file type is not recognized or the file content exceeds the parser's processing capacity.
- In advanced orchestration, the AI conversation node following a decision node does not receive the user's initial question. This happens when the decision node's workflow design does not explicitly pass the user input as preceding context to the subsequent AI conversation node.
- Conversation logs only display system-generated identifiers, without providing actual user identity information. This indicates that the integrating party did not pass the user identifier field in the FastGPT conversation interface or did not configure the corresponding user identity mapping.
How to Verify the Configuration
- Upload medical documents of different types (e.g., plain text, PDF, images) and varying sizes to check if they are successfully parsed and indexed by the knowledge base.
- Conduct multi-turn simulated consultations. Observe whether the conversation model accurately understands the context, cites knowledge base information, and correctly handles key medical terms and numerical values.
- In the conversation logs, verify that user identity information is correctly recorded, and that the input, output, and cited knowledge entries for each conversation align with expectations.
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.