What Data for This Category Looks Like
This category of data primarily comes from cell capacity separation data, manufacturing execution system (MES), welding equipment, torque tools, BMS software version repository, and EOL test benches. Daily documents include cell grouping rules, bill of materials (BOM), welding and torque operation records, BMS flashing logs, insulation and voltage withstand test reports, EOL test data, and system configuration lists. Data updates once per produced battery pack or energy storage cabinet, and updates synchronously with BMS software version changes. Data fields include cell parameters, assembly torque values, test voltages, software version numbers, and other items. There is no unified cross-manufacturer format, so corresponding fields must be extracted from production documents for a single batch.
Constraints Imposed by These Characteristics on Multi-turn Dialogue and Prompt Engineering
Data for this category is scattered across multiple types of production equipment and systems, with high update frequency for single-batch data. Multi-turn dialogue must support context association across data sources, to ensure the latest batch documents are called for every query. Production documents of different types have large differences in their fields. Prompts must clearly specify extraction rules for each document, to avoid mixing up parameters from different documents. Bid document and quotation preparation requires verification of multi-dimensional parameters. Multi-turn dialogue must retain historical verification records, to prevent errors caused by repeated input of the same information. Additionally, data has no unified format. Prompts must adapt to the field structure of single-batch documents, to ensure extracted content meets bid requirements.
Configuration Settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Dialogue history retention duration | 7200 seconds | Covers the full cycle from demand submission to finalization for single-batch battery pack bids |
| Context window size | 8000–12000 token | Adapts to content loading of multiple production documents, BOMs, and test reports |
| Variable binding rules | Associate variables by production batch | Ensures each call extracts exclusive production data for the corresponding batch |
| Multi-turn dialogue history call | Enable messages history records | Retains each round of parameter verification process to avoid repeated input of the same information |
| Dialogue ID management | Generate chatId based on bid project ID | Distinguishes dialogue content for different bid projects to facilitate subsequent archiving and querying |
| Prompt template update | Update synchronously with BMS version changes | Adapts to latest software configuration parameter requirements |
Three Common Misconfigurations
- Symptom: After configuring multi-turn dialogue, corresponding dialogue records for
chatIdcannot be retrieved in the database. Cause: A uniquechatIdis not generated based on bid project ID, leading to incorrect archiving of dialogue records and inability to query by project dimension. - Symptom: Knowledge base search works normally during debugging, but no knowledge base matching content is returned during official AI dialogue. Cause: The prompt does not clearly associate production documents bound to
variablesas a reference basis, leading to failure to load corresponding data in the dialogue context. - Symptom: Results generated by the program execution module cannot be passed to the AI dialogue module as background knowledge. Cause: The program output results are not bound to the
variablesvariable, leading to inability to access this content in the dialogue context.
How to Verify Proper Configuration
A query targeting single-batch production parameters is executed. The dialogue interface is checked for retained messages history records, and each round of input and output is confirmed to be correctly stored. chatId is checked for match with the current bid project ID, and dialogue records for different projects are confirmed to not be mixed. The program execution module is called to generate test data. The test data is bound to the variables variable, then a query is initiated. The AI dialogue is confirmed to correctly reference the test data. The context window size is adjusted. The number and content of loaded production documents are verified to meet current bid requirements, with no truncation or omission.
The values provided here are common starting points, and should be evaluated against samples relevant to the specific deployment.