What this category of data looks like
Data sources for this category include wafer test maps, MES systems, bonding and molding equipment, ATE test machines, X-ray/SAM inspection equipment, and customer specification documents. Daily documents include packaging process cards, bonding and molding parameter tables, incoming wafer maps, test program version lists, yield reports, and failure analysis reports. Data updates occur per production batch, per chip test, and once daily. Data fields include wafer number, batch number, equipment operating parameters, test bin codes, and failure types. Some parameters use Pascals and milliseconds as units.
What constraints these characteristics impose on workflow orchestration
Multi-source heterogeneous data characteristics of this category require workflows to be configured with cross-data source call nodes. These nodes integrate equipment operating data, MES production data, and inspection report data. High-frequency production update rhythms require workflows to support both event-triggered and batch-triggered modes. These modes adapt to different scenarios of single-chip testing and batch production. Dedicated fields for professional process parameters and failure codes require AI chat nodes in workflows to be bound to the knowledge base of the corresponding category. This ensures Q&A matches industry terminology. Format differences across different documents require data processing nodes in workflows to be configured with unified field mapping rules. This avoids process interruptions caused by format incompatibility.
How to Set Configurations
All following configuration items are compatible with FastGPT v4.17.0.
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
| Question Optimization | Retain core process terminology, filter redundant expressions | Professional terminology in semiconductor packaging and testing must be accurately conveyed to avoid Q&A ambiguity |
| Maximum Context | 800–1200 characters | Process documents and test data for this category are lengthy, so sufficient context is needed to match professional content |
variables | Bind batch number, equipment model, and test bin code variables | Process parameters and inspection data related to batches must be dynamically called during the production process |
| Node Trigger Condition | Trigger based on inspection events reported by equipment | Equipment inspection and maintenance Q&A must match real-time equipment operating status |
| Code Run Node Processing Rules | Filter think tag content, only retain business logic output | Non-essential thought processes must be hidden, only maintenance suggestions or inspection results should be displayed |
nodeId | Configure a unique identifier for each process node | Facilitates workflow debugging and abnormal node location, and adapts to multi-process workflow orchestration |
Three Common Mistakes
- Phenomenon: The think tag content of the AI chat node remains displayed after the workflow runs. Cause: The code run node is not correctly bound to the output link of the AI chat node, or the filtering rule does not cover all output fields.
- Phenomenon: The question optimization configuration cannot load historical session records. Cause: The context storage switch for the workflow is not enabled, or
variablesis not associated with the session ID parameter. - Phenomenon: After restoring a project backup, the background knowledge base and agent do not display the corresponding content. Cause: The backup file only stores project configurations, and associated knowledge base files are not synchronized. Knowledge base data must be imported separately.
How to Confirm the Configuration is Complete
- Trigger a simulated inspection event once. Verify the workflow executes in the configured node order, and that the
variablesparameter correctly passes the batch number and equipment model. - Run a workflow that includes a code node. Confirm no think tag-related text appears in the output content.
- Review the workflow's node logs. Confirm each node corresponding to
nodeIdhas a correct execution record. - Upload a test file conforming to the format of this category. Verify the HTTP interface can normally receive and transfer the file to subsequent nodes.
The configuration values provided here are common starting points, and should be evaluated against the reader's own samples.