What this category’s data looks like
Business data for electric motor and generator manufacturing comes from electromagnetic design data, copper wire and insulating material batch records, winding equipment operation logs, dynamic balancing machine testing data, electrical test bench reports, and ERP/MES systems. Data is updated once per motor or per product batch, and synced whenever a formal design change is made. Daily documents include winding process cards, insulation treatment records, dynamic balancing reports, withstand voltage insulation test records, formal and factory test reports. Core fields include winding turns, insulating material batch numbers, test voltage, and temperature rise values. Units include turns, batch numbers, kV, K, and other standard units.
What constraints do these characteristics impose on the "multi-turn dialogue and prompt engineering" link?
The above data characteristics impose three core constraints on multi-turn dialogue and prompt engineering configuration. First, data is updated per unit or per batch. The chatId must be used to bind the corresponding batch or equipment ID to avoid mixing data across batches. Second, documents include exclusive fields such as winding turns and test voltage. Prompts must preset variables extraction rules to ensure the model can accurately locate core business data for this category. Third, data sources cover multiple types of equipment and systems. Multi-turn dialogue must gradually confirm information such as material batches and test results to prevent missing key content that would affect the accuracy of tender documents and quotations.
How to configure the settings
The following configurations are developed based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chatId | Bind to motor batch number or final assembly equipment ID | Ensures uniqueness of data per unit or batch, avoids mixing data across batches |
variables | Preset winding turns, insulating material batch, and test voltage as extraction variables | Matches core fields of category documents, facilitates subsequent quotation and document calls |
messages | Only pass single-batch associated data per round of dialogue | Controls context scale, avoids redundant information interfering with model output |
| Context window upper limit | 8000–12000 token | Calibrated through actual testing, adapts to the text length of winding process cards and test reports |
| Variable extraction matching threshold | 0.85 | Ensures accuracy of extracted fields such as winding turns and test data |
| Language control rule | Output in the language specified in the prompt | Matches the language requirements of tender documents, avoids output content inconsistent with the prompt |
Three common misconfigurations
- Symptom: After passing the
variablesparameter via API, the model fails to extract winding turns data for the corresponding motor batch, and returns content unrelated to the target category. Cause: Thevariablesparameter is not associated with the batch data bound tochatId, causing variables to not map correctly to the corresponding production documents. - Symptom: The text content extraction node returns empty fields. After triggering a specified reply, the workflow terminates directly without waiting for new user input. Cause: The user input listening logic for multi-turn dialogue is not configured, and the node setting for waiting for user interaction is not enabled.
- Symptom: The prompt specifies English for output, but the model only outputs Chinese. Cause: The language rule is not clearly defined in the prompt, or Chinese historical dialogue data is mixed in the context window, interfering with model output.
How to verify that the configuration is complete
- Pass the preset
variablesparameter and the corresponding batchchatId, check whether the model output includes preset fields such as winding turns and insulating material batch. - Trigger the text content extraction node, input a known motor batch number, confirm that the returned fields match the content of the corresponding production documents.
- After configuring the language control rule, input an English prompt, check whether the model output language matches the prompt requirements.
- Upload a category-specific withstand voltage insulation test report file, confirm that the parsing node can correctly extract core data such as test voltage and temperature rise values.
The values provided are common starting points and should be measured against the reader's own samples.