What the data for this category looks like
Financial report data for biologics enterprises mainly comes from periodic reports and temporary announcements disclosed by domestic and overseas stock exchanges, as well as R&D pipelines and batch issuance information officially released by enterprises. Update schedules follow regulatory requirements: quarterly reports are disclosed within 45 days after the end of the quarter, annual reports within 120 days after the end of the year, and temporary announcements are updated in real time alongside major events. Document structures include consolidated financial statements, main business revenue breakdowns (listing specific revenue amounts by subcategories such as vaccines, blood products, monoclonal antibodies, etc.), R&D investment details, clinical pipeline progress, management discussion and analysis, and other modules. Fields cover revenue amounts, batch issuance quantities, single-dose pricing, R&D stages, clinical time nodes, and more. Common units are RMB yuan, ten thousand doses, ten thousand yuan, and date formats.
What constraints these characteristics impose on multi-turn conversation and prompt configuration
The scattered sources and complex structure of biologics financial report data create multiple constraints for multi-turn conversation and prompt setup. First, update schedules vary significantly across different data sources. Static data from periodic reports and dynamic data from batch issuance and temporary announcements must be distinguished during conversations. Prompts must clearly specify the time range and type of data sources. Second, fields such as revenue breakdowns by subcategory and R&D pipelines require precise positioning. Prompts must limit extraction scopes to avoid confusing business data from different biologics subcategories. Third, unit differences across multiple fields require unified calibration during conversations. Prompts must explicitly require output to include corresponding units. Finally, cross-module business associations, such as clinical progress and corresponding revenue contributions, require retained context memory. Multi-turn conversations must be configured with sufficient context window length.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 8000–16000 characters | Biologics financial reports contain multi-module detailed content. Multi-turn conversations need to retain multi-round business-related context to avoid context overflow |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Financial report documents include detailed revenue breakdowns and R&D pipeline details, which are long-form content. Parsing takes a long time, so the timeout period must be extended |
Recall count | Top 8–10 entries | Biologics financial reports have many detailed fields. Sufficient detailed data must be recalled to support accurate conversations |
Similarity threshold | 0.75–0.85 | Industry-general non-financial report data must be filtered to accurately match exclusive fields and content of biologics financial reports |
Workflow global variables | Configured by data source category | Biologics financial report data comes from multiple sources such as periodic reports, batch issuance, and R&D pipelines. Parameters need to be passed by category |
UPLOAD_FILE_MAX_SIZE | 500 MB | Annual financial report documents include complete business details, resulting in large single-file size. The upload limit must be adapted |
The parameter values provided on this page are conventional recommendations used as a starting point for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require specific analysis. It is recommended to test on your own samples before finalizing settings.
Three common mistakes
- Phenomenon: After the workflow runs, the input fields of the AI conversation node are empty, and the output content of the code running node cannot be read. Cause: The output result of the code node is not bound to the input parameters of the conversation node, and the variable transfer rules of the workflow are not configured.
- Phenomenon: The revenue data output in multi-turn conversations includes business amounts from non-target categories, or is not split by specified subcategories. Cause: The prompt does not clearly define the extraction range of biologics subcategories, and does not associate with the detailed fields of the corresponding financial report documents.
- Phenomenon: After upgrading to version v4.8.10, a query containing only a dozen words receives a full answer output in one go, without streaming return. Cause: The
stream_responseparameter is not configured to enabled, or the context window setting exceeds the model's supported range, resulting in early termination of streaming output.
How to confirm correct configuration
- Upload a single annual financial report document, check whether the knowledge base parsing results include target fields such as detailed revenue breakdown and R&D pipeline, and confirm that the parsing configuration meets business requirements.
- Initiate a test conversation with multiple rounds of follow-up questions, such as first asking about total revenue for a quarter, then asking about the revenue amount of vaccine business, and check whether the conversation context is correctly retained and the output result is associated with the corresponding detailed data.
- Run a workflow including a code node and a conversation node, check whether the conversation node can correctly read the output content of the code node, and confirm that the variable transfer configuration is correct.
- Initiate a short text test query, check whether the answer uses streaming output, and confirm that the streaming output configuration is enabled as required.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.