What Data for This Category Looks Like
Data sources for this category include product drawings and 3D models, CAPP/CAM systems, machine tool collection terminals, tool management systems, MES systems, and measurement and inspection tool data. Data updates are triggered for each work order, single workpiece machining run, sampling inspection step, or tool change. Daily documents cover process routes, operation cards, tool lists, CNC programs, first-piece inspection records, patrol inspection forms, and measuring tool calibration records. These documents include fields such as process number, cutting parameters, tool number, and dimensional deviation values. Some documents use specialized units such as micrometers and minutes.
Constraints on Multi-turn Dialogue and Prompt Engineering for This Category
The multi-source, high-frequency updated data characteristics of this category require multi-turn dialogue to bind chatId to track the complete audit process of a single work order, avoiding context confusion. Specialized fields and units for different document types require prompt engineering to clearly specify extraction targets, document types, and field rules. For example, only extract cutting parameters from operation cards, and do not associate irrelevant fields from other documents. High-frequency updated tool and quality inspection data require dialogue context to limit valid rounds, preventing outdated data from interfering with audit judgments. Differences in the code format of CNC programs and the calibration record format of measuring tools also require prompt engineering to adapt to the parsing logic of different data types.
Configuration Settings
These configurations are based on FastGPT v4.17.0:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chatId | Bind the unique identifier of the current work order | Ensure multi-turn dialogue is associated with the same audit process, avoiding context interference |
messages | Retain the most recent 8 rounds of dialogue content | Adapt to high-frequency updated quality inspection and tool data, preventing outdated information from interfering |
variables | Pre-set audit variables such as work order number and tool model | Quickly match the process and quality inspection documents corresponding to the workpiece |
| Single search limit | Top 6 recalled results | Match the retrieval volume of multiple types of documents in this category, avoiding redundant data |
| Maximum context length | 4000–6000 characters | Adapt to the content length of single documents such as operation cards and CNC programs, calibrated based on actual testing |
| Text extraction trigger rule | Match corresponding fields by document type | Adapt to the specialized field formats of different documents in this category |
Three Common Configuration Mistakes
- Phenomenon: Empty fields are returned during the text extraction phase, and the system triggers the specified response without entering a state waiting for user input. Cause: No dialogue wait trigger rule is configured, causing the process to terminate directly and unable to receive supplementary user information.
- Phenomenon: After configuring an English prompt, the model only outputs Chinese content. Cause: No mandatory output language rule is specified in the prompt, and no language matching configuration for the English knowledge base is associated.
- Phenomenon: Parsing errors occur after uploading CNC program or operation card files. Cause: No custom parsing rules are configured for the specialized document formats of this category, such as G-code and dimensional parameters with units, causing the general extraction logic to fail to adapt.
How to Verify Successful Configuration
- Initiate a simulated work order dialogue, input keywords for different types of documents, and verify whether the dialogue context is retained bound to
chatId. - Configure an English prompt and input an English query, and verify whether the model output meets the specified language requirements.
- Upload typical documents of this category, such as operation cards and CNC programs, and verify whether text extraction can correctly return preset fields.
- Trigger a text extraction failure scenario, and verify whether the system triggers the specified response and waits for user supplementary information.
The values provided are common starting points and should be measured against the reader's own samples.