What the data for this category looks like
The main entities of this category's data are skincare, haircare, makeup, and special cosmetics manufacturing enterprises. Daily documents include formula work sheets, weighing and feeding records, emulsification or filling records, cleaning and site clearance records, semi-finished and finished product inspection reports, and label review materials. Data is sourced from registered and filed formulas, raw material safety documents, production equipment, ERP/MES systems, laboratory data, and packaging material version files. Update frequency follows per production batch or each line change. Data fields must align with regulatory filing requirements, including restricted substance information for raw materials, production parameters, batch numbers, and claimed content, among other details.
What constraints these characteristics impose on the tool calling and plugins workflow
The batch or line change based data update pattern of this category requires tool calling to be triggered tied to the production batch dimension, to avoid cross-batch data confusion. Multi-source data is scattered across ERP/MES, production equipment, laboratory systems, and packaging material management repositories. Plugins must support connecting to multiple data source interfaces to uniformly pull raw material, production record, and inspection data for the corresponding batch. Document fields must match regulatory filing requirements. Tool calling must verify whether raw materials appear on the restricted substance filing list, whether formula parameters match registered and filed content, and also link packaging material versions to filed claimed content to ensure label review data matches.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Tool calling trigger timing | Within 10 minutes after production batch creation | Matches the data update rhythm of per-batch production or line changes, ensuring the latest batch data is used for calls |
| Workflow/API tool selection | auto | Automatically adapts to multi-source data verification scenarios, no manual tool specification required, supports diverse audit needs for raw materials, formulas, and labels |
| Multi-data source pull limit | Top 8 pieces of data from the same batch | Covers necessary data for core links such as feeding, emulsification, inspection, and site clearance involved in single-batch production |
| Single tool call timeout | 600 seconds | Adapts to the time required to pull data across ERP, production equipment, and laboratory systems, prevents call interruptions due to large data volume |
| Workflow/API function-call configuration | Enabled | Supports custom tool calling logic, adapts to complex raw material restricted substance verification scenarios |
| Result verification matching rule | Based on actual measurement calibration | Adjust matching precision for different verification scenarios including formula consistency, raw material restricted substances, and label claims |
Three Common Errors
- Phenomenon: After calling a tool, the response returns "No data associated with the current batch found". Cause: Tool calling is not bound to the production batch, and historical data from non-current production batches is pulled, resulting in failure to match feeding, formula, and other information required for current audits.
- Phenomenon: The style of the FastGPT page embedded via iframe does not match the project, and global styles cannot be modified. Cause: Custom style isolation configuration is not enabled, and default iframe embedding is used directly, leading to style conflicts.
- Phenomenon: Calling the session interface does not return the expected audit result, and the prompt word does not take effect. Cause: The prompt word parameter of the session interface is not configured correctly, and the scenario requirements of current batch audit are not matched.
How to Confirm Proper Configuration
- Manually trigger a tool call, check whether the returned data includes core document content such as formula, feeding, and inspection for the currently selected production batch.
- Adjust the tool call timeout to simulate cross-system data pulling scenarios, confirm that the call is not interrupted due to timeout.
- Submit a set of test data known to meet filing requirements, check whether the tool verification result matches expectations, and adjust the verification rules based on the result.
- Check whether the
toolChoiceandfunctionCallconfigurations are correct, confirm that the tool can automatically adapt to multiple types of audit scenarios without manual tool switching. - Confirm that the front-end embedded style configuration is enabled, view the page style through browser developer tools to confirm it meets project requirements.
The values listed here are common starting points, and should be evaluated against the reader's own samples.