Data profile for this category
Data for this category originates from animal quarantine records, supplier batches, production traceability logs, temperature collection data, metal detection results, and laboratory test reports. Daily documents include quarantine certificates, slaughter and cutting records, ingredient marinating logs, thermal processing records, cold chain temperature logs, and microbial test reports. Data updates occur per batch and per shift. Cold chain temperature data is collected continuously. Data fields include batch unique identifiers, quarantine numbers, temperature values, test results, workshop numbers, and additional fields. Units include degrees Celsius, batch numbers, test dates, and others.
Constraints on multi-turn dialogue and prompt engineering
The binding of batch and quarantine information for this category requires multi-turn dialogue to consistently carry batch unique identifiers. This prevents context confusion between data from different production batches. The real-time nature of continuous temperature control data requires dialogue to support quick retrieval of the latest collection records. Prompt engineering must explicitly specify temperature thresholds and exception determination rules. Requirements for separating raw and cooked product zones and controlling sterilization temperatures require prompts to distinguish document fields across different processes. This avoids mixing up data related to cross-contamination risks. The need to correlate data from multiple sources (quarantine, testing, temperature) requires multi-turn dialogue to integrate fields from different sources. This ensures complete and accurate information in exception reports.
Configuration settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chatId | Bind to production batch unique identifier | Ensure dialogue context for each product batch is linked to its exclusive quarantine and production records |
messages | Retain the most recent 20 turns of dialogue context | Cover the full process dialogue for a single product batch from slaughter to testing, and prevent loss of critical context |
variables | Pre-set batch number, quarantine number, and temperature threshold variables | Quickly access category-specific data fields, and reduce repetitive input operations |
| Dialogue timeout | 600 seconds | Match the continuous reporting interval of cold chain temperature collection, and avoid interrupting real-time dialogue due to timeout |
| Maximum context token count | 8000–12000 tokens | Accommodate multiple document contents for a single product batch, and meet the context window requirements of large language models |
Three common configuration mistakes
- Symptom: A connection error prompt appears after submitting a dialogue, and recovery fails even after enabling a network proxy. Cause:
variablesare not configured correctly to bind access permissions for the enterprise intranet data interface, leading to network request blocking when the large language model calls data sources. - Symptom: Generated exception reports do not include quarantine certificates or cold chain temperature curve image links stored in the knowledge base. Cause: Prompts do not explicitly require association with image documents in the knowledge base, and rules for triggering knowledge base image retrieval are not configured.
- Symptom: Extracted keyword variables fail to accurately match category-specific fields, such as unable to correctly identify carcass numbers or slaughter workshop numbers. Cause: Category-specific fields are not explicitly designated as extraction targets in prompts, causing generic variable extraction logic to fail to adapt to business requirements.
How to verify correct configuration
- Initiate a test dialogue that includes a production batch number, and check if the returned context includes exclusive data fields for that batch.
- Trigger an exception report generation request, and verify that the returned content includes preset
variablessuch as quarantine numbers and cold chain temperature thresholds. - Simulate a network interruption then resume the dialogue, and check that the conversation resumes normally without connection error prompts.
- Test the keyword extraction function, and confirm that category-specific business fields such as carcass numbers and workshop numbers are accurately extracted.
The values provided are common starting points and should be measured against the reader's own samples.