What data for this category looks like
This category of data typically originates from ERP orders and inventory, BOMs and process routes, equipment capacity, staff shifts, work-in-progress, and supplier deliveries. Daily processed documents include sales orders, master production schedules, material requirements plans, weekly and daily scheduling sheets, work orders, material shortage lists, and delivery change notices. Data update cycles are daily, weekly, or triggered in real time alongside orders or abnormal events. Document fields include order numbers, material codes, process man-hours, equipment numbers, shift time periods, work-in-progress quantities, and more. Units include pieces, hours, units, days, and others.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The multi-source and heterogeneous data characteristics require the retrieval link to adapt to different document field formats, to avoid recall result deviations caused by field mismatches. The high-frequency update feature requires the recall logic to support real-time or near-real-time synchronization of the latest scheduling, work order, and abnormal data, to ensure retrieval results align with current production status. The diverse document types cover structured forms and unstructured instructions, so retrieval matching rules need targeted adjustments, prioritizing content strongly related to production planning and resource constraints. Additionally, the core requirement of production scheduling is to accurately match process, equipment, and material information; recall results must be strictly linked to the specific parameters of the current work order, to avoid irrelevant data interfering with scheduling decisions.
How to configure
FastGPT v4.17.0 supports adjusting knowledge base retrieval and recall logic for production planning and shop scheduling scenarios via the following configuration items:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Matches the precise field association requirements of production plans, avoiding recall of low-relevance non-scheduling documents |
limit | Top 8 entries | Controls the number of recall results per request, adapting to the limited context length requirements of production ledger summary |
searchMode | mixedRecall | Combines keyword and semantic retrieval, covering structured work order fields and unstructured scheduling instructions |
usingReRank | Enabled | Performs secondary sorting on recall results, prioritizing content strongly related to current scheduling tasks |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval keywords, covering synonymous expressions of core fields such as work order numbers and material codes |
collectionIds | Bind to dedicated datasets for production planning and shop scheduling | Limits retrieval scope, avoiding mixing non-production data from other categories |
Three common misconfigurations
slow operation xxxxmslogs appear continuously during retrieval after deployment. The cause is failure to limit the retrieval dataset scope, leading the system to traverse all knowledge base files and exceed the MongoDB response threshold.- Some files are written directly as raw text without generating question-answer pairs. The cause is failure to set dedicated chunking rules for structured scheduling sheets and work orders, so the system cannot identify the semantic boundaries of structured fields.
- The
insufficient_quota Current group upstream load is saturated, please try again latererror is returned during retrieval. The cause is failure to configure reasonable retrieval parameters, with too much context carried in a single recall request, triggering platform quota limits.
How to confirm the configuration is properly set
- Manually upload a test work order document, execute retrieval, and check whether the fields of the recall results match the test document.
- Check the retrieval logs to confirm that
slow operationerrors no longer appear, verifying that the dataset scope limitation has taken effect. - Trigger a ledger summary task, and check whether the number of recalled documents meets the set retrieval upper limit requirement.
- Submit a query covering abnormal scheduling scenarios, confirming whether the re-ranking function prioritizes recalls of relevant material shortage lists and delivery change notice documents.
The values provided are common starting points and should be measured against the reader's own samples.