What this category’s data consists of
Printed circuit board factory engineering, production, quality control, and laboratory teams generate this category of data. Daily documents include engineering production instructions, Gerber analysis records, drilling and lamination parameters, chemical solution analysis reports, AOI inspection reports, electrical testing reports, and slice inspection reports. Data sources include customer Gerber files, drill tapes, BOM lists, CAM files, equipment operating parameters, tank solution test results, data collected by AOI and electrical testing equipment, and laboratory test records. Updates occur per production batch, per shift, or via continuous collection. Fields include process parameters, test results, equipment operating status, and other data that changes dynamically with production stages.
What constraints these characteristics impose on multi-turn dialogue and prompt engineering
Data sources for this category are scattered, covering multiple document types such as engineering drawings, process parameters, and test reports. Multi-turn dialogue relies on clear prompt engineering to distinguish call priorities for different documents, and avoid confusion between process requirements of different stages. Data update frequency changes dynamically with production batches and shifts. Dialogue context must be limited to valid data for the current production batch, to prevent errors caused by calling expired parameters. Equipment inspection and maintenance carry risks of full batch scrapping. Multi-turn dialogue retains historical interaction records, to track continuous changes in abnormalities and assist in locating root causes of faults.
The following configurations are based on FastGPT v4.17.0.
How to set configurations
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
chatId | Bind to the current production work order ID | Distinguish dialogue context across production batches, avoid mixing data from different work orders |
messages | Retain the last 10 rounds of interactions and current production batch data | PCB process abnormalities require tracking multi-step troubleshooting processes, avoid context overflow |
variables | Pass in the Gerber number and equipment model of the current work order | Accurately match corresponding production data and equipment inspection records |
| Maximum context length | 800–1200 characters | Balance information capacity per dialogue turn and response speed, adapt to the text length of PCB process parameters |
| Input length interception | 1500 character threshold | Prevent errors triggered by exceeding the AI model's input limit, align with actual user input scenarios |
Three common configuration mistakes
- Phenomenon: The dialogue request interface returns results without a cite reference ID field. Cause: The source data association configuration for dialogue context is not enabled, and the reference identifier of the corresponding document is not bound.
- Phenomenon: An interface error is triggered after input text exceeds the preset threshold, with no advance interception prompt. Cause: No input length verification rule is configured, and the request is submitted directly to the AI model.
- Phenomenon: JPG format equipment inspection photos cannot be directly accessed. Cause: No pre-processing node for file-to-text conversion is added, and only plain text input channels are enabled.
How to confirm configuration is complete
- Initiate a test dialogue that includes the Gerber number and equipment model, confirm that the
variablesparameter is correctly passed, and that the dialogue context can associate corresponding production data. - Input test text that exceeds the preset threshold, confirm that the system intercepts in advance and prompts that the input exceeds the limit, and no interface error is triggered.
- Upload a JPG format inspection photo, confirm that the system can complete parsing and include the text content in the dialogue context.
- Initiate a multi-round troubleshooting dialogue, confirm that
chatIdis bound to the current work order ID, and that historical interaction records can be retained and called normally.
The values provided are common starting points and should be measured against the reader's own samples.