What Data for This Category Looks Like
Typical data for this category comes from manufacturing enterprise PMC, planning, shop scheduling, and material control departments. Daily generated documents include sales orders, master production schedules, material requirement plans, weekly and daily production schedules, work orders, material shortage lists, and delivery change notices. Most data is sourced from ERP orders and inventory, BOMs and process routes, equipment capacity, staff shifts, work in progress, and supplier delivery information. Update frequency is typically daily, weekly, or real-time when orders or exceptions occur. Documents often include fields such as order number, material code, process name, equipment model, capacity value, and delivery deadline. Units include pieces, hours, units, and days, among others.
Constraints on Knowledge Base Retrieval and Recall
This category’s data includes multi-dimensional production and order-related information, with most fields being structured numerical values and codes. Retrieval must precisely match business fields to prevent irrelevant documents from being included. High-frequency real-time updated data creates pressure for knowledge base incremental synchronization. Dynamic update-adaptive recall logic is required to ensure recalled content reflects the latest capacity, scheduling, and order information. Field formats vary significantly across different documents. For example, production schedules and work orders have different field dimensions. Unified recall and integration of multi-document structures must be supported. The bidding and quotation scenario requires quick association with historical orders and capacity data, which has clear requirements for retrieval response speed. The retrieval scope must be limited to category-related business data.
Configuration Settings
The following configurations are set based on FastGPT v4.17.0:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | This category’s data mostly uses structured fields. This range filters low-match non-business documents while balancing recall accuracy and coverage |
limit | Top 8–12 entries | The bidding and quotation scenario requires quick integration of limited core production data. Excessive recall will exceed the context window limit |
searchMode | embedding | This category’s data contains many numerical values and codes. Vector search accurately matches production planning and order information related to business needs |
usingReRank | Enabled | After multi-document recall, results must be reordered by business relevance to prioritize capacity and scheduling data matching the current bidding order |
datasetSearchUsingExtensionQuery | Enabled | Queries can be expanded based on bidding order keywords to supplement implicitly associated data such as material requirements and equipment capacity |
collectionIds | Bind dedicated knowledge base collections for production planning and shop scheduling | Prevent non-production documents from being recalled, ensuring the retrieval scope only covers category-related business data |
Three Common Configuration Errors
- Exceeded the associated knowledge base limit per conversation: The interface returns the error "Associated knowledge bases have reached the upper limit". This occurs when dedicated knowledge base collections are not planned in advance, leading to quota exhaustion from mixing irrelevant documents.
- Recalled result total word count exceeds the context window: Returned results are truncated or complete quotations cannot be generated. This happens when a reasonable
limitparameter is not set, with excessive recalled documents causing context overflow. - Chaotic recall result ordering: Returned scheduling data is irrelevant to the current bidding order. This occurs when
usingReRankis not enabled, so recalled results are not reordered by business relevance.
How to Verify the Configuration is Correct
- Check knowledge base collection binding: Confirm only knowledge bases related to production planning and shop scheduling are bound, with no irrelevant documents mixed in.
- Test retrieval matching accuracy: Enter an order number to verify recalled results include fields such as material requirements and equipment capacity corresponding to that order, and that filtered results meet the
similaritythreshold requirement. - Verify reranking function: Enter a query for an order insertion scenario to confirm recalled results prioritize the latest equipment shift and capacity data.
- Confirm extended query is active: Enter a bidding delivery deadline keyword to verify relevant documents such as delivery change notices and weekly and daily production schedules are automatically associated.
The values given are common starting points and should be measured against the reader's own samples.