What This Category of Data Looks Like
Daily documents for this category include sales orders, master production plans, material requirement plans, weekly and daily production schedules, work orders, material shortage lists, and delivery change notices. Data sources come from ERP orders and inventory, BOMs and process routes, equipment capacity, staff shifts, work-in-progress, and supplier deliveries. Update frequency is typically daily, weekly, or triggered by orders or exceptions. Most documents contain fields such as order numbers, material codes, process numbers, capacity values, shift times, and work-in-progress quantities, with units including pieces, hours, days, and others.
Constraints Imposed on Vector Models and Indexes
The diverse data sources and frequent updates of this category require indexes to support incremental synchronization and scheduled refreshes, matching the daily and exception-triggered update rhythms. Most documents are semi-structured content spliced from multiple fields, so splitting must avoid damaging business association logic, and targeted chunking rules must be set. There are many fields and cross-document associations, so vector models must adapt to multi-dimensional semantic splicing, while indexes must support quick recall by associated fields. In addition, documents such as schedules and work orders vary widely in length, so sufficient context processing space must be reserved.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapt to the text length of complete business units such as work orders and production schedules, avoiding splitting that damages the business logic of associated fields like processes and materials |
| Overlap Length | 100–150 characters | Cover cross-chunk associated fields, ensure the integrity of context semantics, and avoid business segment breaks |
maxToken | Set to the selected embedding model/service input limit | This is the configured embedding model's maximum token count, not retrieval or LLM context. Check the selected model or provider documentation. See the v4.17.2 model-field reference |
defaultToken | Calibrate with representative work orders and schedules | This is the default token count used for direct text splitting; inspect sample chunks. See the v4.17.2 model-field reference |
| Similarity Threshold | 0.72–0.78 | Distinguish semantic differences between similar scheduled work orders and abnormal work orders, accurately recalling historical data related to quality problems |
| Single Search Limit | Top 8 entries | Shop scheduling data is associated with multi-dimensional information, and a small number of highly relevant results can support the business needs of quality traceability and defect analysis |
Three Common Mistakes
- Phenomenon: The imported dataset.csv only contains the index field and no content field. Cause: Business fields of documents such as work orders and production schedules were not correctly spliced, and only index metadata was exported without associating actual business text content.
- Phenomenon: The local bge-large deployment reports insufficient GPU memory. Check the model/runtime memory requirements, available device memory, concurrency, and runtime logs.
defaultTokencontrols text splitting; it does not configure model parallel loading. - Phenomenon: The custom index cannot match the fields of the corresponding MongoDB table. Cause: Associated fields of documents such as production schedules and work orders were not mapped to the metadata fields in the index configuration, resulting in a mismatch between the index and database table fields.
How to Confirm Proper Configuration
- Upload a test weekly production schedule, check whether the parsed text chunks retain complete process, capacity, and shift information, with no truncation or splitting errors.
- Run a quality traceability query, verify that the returned results include the associated documents of the corresponding work order, and that the fields match the preset business logic.
- Measure resource usage on the target hardware and inspect model/runtime logs while indexing representative data; requirements depend on the selected model, runtime, and concurrency.
- Check the index update logs of FastGPT v4.17.2 and confirm that scheduled or exception-triggered schedule data is synchronized as intended.
The values provided here are common starting points and should be evaluated against the reader's own samples.