Multi-turn Conversations and Prompts for Health Management

Health Management Data Characteristics

Health management product data originates from user health records, physical examination reports, wearable device data, dietary records, medication history, and health questionnaires. Data update frequencies vary. Physical examination reports are typically annual, while wearable device data can be real-time or daily. Document structures often mix structured data with unstructured descriptions, such as blood test results and imaging report narratives. Dietary records may include food names, intake amounts, and nutritional components. Fields include physiological indicators like blood pressure (mmHg), blood glucose (mmol/L), heart rate (bpm), and medication dosages (mg), and exercise duration (min).

Constraints on Multi-turn Conversations and Prompts

Frequent updates and diverse sources of health management data require efficient data ingestion and update mechanisms for the knowledge base. This ensures the timeliness of conversational content. The mix of structured and unstructured data in physical examination reports means document chunking must balance precise numerical matching with semantic understanding of descriptive text. Physiological indicators and medication dosages, with specific units, require accurate identification and conversion in multi-turn conversations to avoid misunderstandings due to unit confusion. Users often ask follow-up questions based on historical health data, such as "What was my blood glucose level during my last physical?" or "Based on my exercise data, what is my calorie expenditure today?". This requires multi-turn conversations to effectively link context and accurately retrieve relevant information from complex data.

Configuration Guide

Configuration Item	Recommended Value	Rationale
`Chunk size` (Chunk Length)	`500–800 characters`	Balances completeness of descriptive text in reports with model context limits
`Recall count` (Recall Count)	`Top 8 entries`	Increases coverage for retrieving multi-source health data
`Similarity threshold` (Similarity Threshold)	`0.75`	Ensures accurate matching for precise physiological and medication information
`Rerank result count` (Reranked Return Count)	`Top 5 entries`	Prioritizes information most relevant to the user's health query
`maxContext`	`4000 token`	Supports longer health history data and multi-turn follow-up questions
`UPLOAD_FILE_MAX_SIZE`	`100 MB`	Accommodates uploading physical examination documents with multiple pages or images

Common Pitfalls

The system fails to parse a physical examination report uploaded by the user during a conversation, displaying "Document too large or format not supported." This usually occurs due to a low UPLOAD_FILE_MAX_SIZE setting or PARSE_FILE_TIMEOUT_SECONDS timing out when processing large PDF files.
In a second follow-up question, the model fails to connect context and provides irrelevant health information. This might be due to an insufficient maxContext setting, leading to historical conversation information being truncated in multi-turn interactions.
When asked "What was my last blood glucose level?", the model provides multiple irrelevant health indicators. This suggests the Similarity threshold (Similarity Threshold) is set too low, or Rerank result count (Reranked Return Count) is too high, leading to the recall of excessive non-critical information.

How to Verify Configuration

Upload a physical examination report containing both structured indicators and unstructured descriptions. Ask questions about specific numerical values and diagnostic results from the report. Check if the replies are accurate and complete.
Engage in multi-turn conversations. For example, first ask "What is my blood pressure history?", then follow up with "What was the most recent blood pressure reading, and how does it compare to the last one?". Confirm the model can connect context and provide effective answers.
Test questions involving physiological indicators with different units. For example, ask "What is my weight in kilograms?", then follow up with "What is that in pounds?". Verify the accuracy of unit conversions.

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.