Multi-Turn Conversations and Prompts for Rare Disease

Data Characteristics

Rare disease registration documents come from diverse sources. These include clinical trial reports, non-clinical study reports, drug manufacturing and quality control documents, and regulatory guidelines. Document update frequencies vary. Clinical trial data typically releases after trial completion. Regulatory guidelines may revise annually or issue supplements irregularly. Document structures are complex. For example, clinical trial reports often contain sections like abstracts, methods, results, and discussions, and may include numerous figures, tables, and appendices. Fields and units are specific. Drug dosages are in milligrams (mg) or micrograms (ug). Study durations are in days (days) or weeks (weeks). Biomarker concentrations are in nanomoles per liter (nmol/L). Statistical indicators like p-value and CI intervals are common.

Constraints on Multi-Turn Conversations and Prompts

The specialized and fragmented nature of rare disease data challenges multi-turn conversation context management. Complex and specialized terminology requires the model to precisely understand user intent. This prevents conversations from deviating due to inaccurate knowledge base recall. Complex document structures mean that simple keyword matching is insufficient for effective knowledge retrieval; stronger semantic understanding is necessary. Frequent regulatory updates demand an efficient synchronization mechanism for the knowledge base to ensure content timeliness. Inconsistent standardization of units and fields across different reports can hinder data extraction and affect prompt generation quality. For example, if a user asks for specific drug pk curve data, the system must extract valid information from non-standardized chart descriptions and present it accurately.

Configuration Settings

Configuration Item	Recommended Value	Rationale
`maxContext`	`2000 characters`	Ensures sufficient context retention in multi-turn conversations to handle the complex terminology and background information of rare disease registration documents.
`Recall Count`	`8 items`	Rare disease knowledge base documents are typically large and highly interconnected. Increasing the recall count helps cover more comprehensive information and avoids missing critical details.
`Similarity Threshold`	`0.78`	Professional terms and concepts in the rare disease field have high distinctiveness. A higher threshold helps filter document segments highly relevant to the user's query, reducing interference from irrelevant information.
`Rerank Return Count`	`4 items`	After recall and reranking, select a small number of the most relevant items to ensure prompt quality and relevance, avoiding overload.
`Segment Length`	`500 characters`	Rare disease documents are dense in content. A moderate segment length helps improve retrieval efficiency and model processing capability while maintaining semantic integrity.
`PARSE_FILE_TIMEOUT_SECONDS`	`600 seconds`	Registration documents often contain large amounts of text and complex formats. Extending the parsing timeout ensures complete processing of large files.

Common Pitfalls

Frequent "I could not find relevant information" or repetitive content in conversations indicates either a Similarity Threshold set too high, filtering out relevant but not exact matches, or a Recall Count set too low, failing to cover enough information.
When users ask about specific drug dosage units, the returned results may have inconsistent or missing units. This occurs if the knowledge base did not standardize and label units like mg and ug during document preprocessing, preventing the model from accurate extraction.
When asked about the latest regulatory updates, the system may respond with outdated information. This happens if the knowledge base update mechanism is not timely, causing the documents fastgpt relies on to lag behind actual updates.

Verification Steps

Select a rare disease registration case with multi-turn Q&A scenarios. Test if the conversation flow consistently understands context and provides coherent, relevant answers.
Ask about clinical data for multiple rare disease drugs with specific dosage and time units. Check if the system's results accurately include unit information and can perform simple unit conversions.
Simulate questions after a rare disease regulatory update. Confirm that the system returns the latest information and cites correct regulatory article numbers.
Query treatment plans or adverse reactions for a specific rare disease. Verify if the system can integrate information from multiple document types (e.g., clinical reports, drug inserts) and form logical answers.

Note: The values provided are common starting points. Measure them against your own samples for optimal configuration.

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.