What this category's data looks like
Undergraduate program revision involves three types of materials: program plans, course syllabi, and credit structure tables. The program plan provides the disciplinary direction and total credit framework, the course syllabus details the objectives and content of each course, and the credit structure table lists the credits and proportions of each module. After these materials enter the model access and configuration stage, the language model handles dialogue and terminology extraction, the indexing model is responsible for multilingual material retrieval, and the reranking model optimizes the ranking of terminology candidates. The material sources are exports from the academic affairs system and submissions from departments; the update cadence changes with revision batches, and the fields include course names, credits, dates, and signatures. Multilingual materials and terminology processing require the same terminology to be consistent throughout the entire text and translated names to be able to refer back to the original text. Therefore, the model binding relationships need to correspond to the material stages, and the terminology glossary and comparison draft are produced after the dialogue and retrieval use this model.
Configuration and implementation checks
This stage is executed by the server-side model channel configuration, outputting model binding relationships and channel responses. The unified terminology glossary and comparison draft are then completed by subsequent AI nodes, code nodes, or manual work. Models are divided into five categories: language, indexing, reranking, speech synthesis, and speech recognition. Different stages bind different model types, and the type labels need to be checked in the model configuration.
| Object | Check | Verification action |
|---|---|---|
| Program plan | Course names, credits, dates, and signatures must be consistent with the course syllabi | Compare the course name and credit fields of the two materials, and check whether the signature fields are complete |
| Course syllabus | The translated name of the same terminology must be able to refer back to the original text and be written consistently throughout | Extract three terminologies, and check the writing throughout and the reference relationship to the original text |
| Credit structure table | The credit amounts and dates of each module must be synchronized with the version of the program plan | Check the credit totals and dates, and inspect the version number and receipt records |
| Language model and indexing model | The model type corresponds to the stage in which it is to be used | Check the type labels in the model configuration, compare them with the model names returned by the channel, and check the values of maxContext and maxTokens |
Three common errors
- Phenomenon: The same course name is written differently in the program plan and the course syllabus. Cause: The two materials were not updated in sync, and the terminology glossary did not cover this entry.
- Phenomenon: The amounts in the credit structure table do not match the program plan. Cause: The versions were not synchronized, and only one material was updated after the revision.
- Phenomenon: Retrieval reports an error related to the indexing model. Cause: The indexing stage is bound to the language model, and the model type does not match the stage.
How to verify the setup
Acceptance for this feature: run one dialogue and one retrieval, check the model identifiers in the responses, confirm that the language model and indexing model are bound correctly respectively, and check the values of quoteMaxToken and toolChoice in the request.
Business test: extract three terminologies and check whether their writing is consistent throughout, verify whether the translated names can refer back to the original text, and record the tokens and array length of the quoted text.
- Check the model type labels in the model configuration, and confirm that language, indexing, and reranking are bound to their corresponding stages.
- Compare the model names returned by the channel with the names listed in the documentation, and check the channel configuration and key source.