Home of the Uedu platformShow a desktop notification when the AI TA finishes replying
Notify me when classmates post messages in the forum
Play an alert sound whenever there is a new notification
Explain how Uedu's AI interactive assignments enable Instructors to design conversational assignments, and use LLM automated marking and instructor review mechanisms to assess Students' dialogue quality.
AI Interact (AI interactive assignment) is the conversational assignment module of the Uedu platform. Teachers can create learning tasks centred on AI dialogue, students complete the assignment through multi-turn conversations with the AI TA, and the system then automatically grades it with an LLM.
The core idea of this module is: conversation itself is the learning process. Through carefully designed System Prompt and assignment modes (mode_code), instructors can guide students through Socratic questioning, role-play, case analysis and other diverse learning activities, and obtain immediate feedback through automatic marking.
Each AI interactive assignment is set up by the Instructor with the following core elements:
| Setting items | Description |
|---|---|
| Assignment mode (mode_code) | Define AI interaction patterns, such as Socratic questioning, role-play, case analysis, and so on |
| System Prompt | AI prompts written by the instructor that define the AI's role, code of conduct and response style |
| Scoring criteria (scoring_criteria) | Multidimensional scoring rules in JSON format (see Section 3) |
| Grading method | use_default_scoring (use platform default) or custom_scoring (teacher-defined) |
The marking criteria are defined in JSON format (scoring_criteria), with each dimension including a name, description and points allocation:
| Field | Type | Description |
|---|---|---|
name | String | Marking dimension name (for example, "Problem depth" or "Logical coherence") |
description | String | Scoring notes and criteria for this dimension |
max_score | Number | Maximum score for this dimension |
use_default_scoring): the platform provides a general multi-dimensional scoring rubric, suitable for general AI dialogue assignmentscustom_scoring): Instructors can customise scoring dimensions and criteria for specific assignment needs, for example by adding a “code correctness” dimension to programming assignmentsAutomatic grading uses the gpt-5-mini model, and response_format=json_object is used to ensure a structured JSON response.
The scoring results returned by the LLM include the following structure:
| Field | Type | Description |
|---|---|---|
criteria_scores | Array | Score and comments for each assessment dimension (score + comment) |
total_score | Number | Sum of scores across dimensions |
overall_comment | String | Overall feedback summarising the Student's dialogue performance |
Students can view the score for each dimension and the specific feedback (comment) provided by the LLM, so they can see where they are doing well or need to improve.
Once AI auto-marking is complete, the Instructor can review the marking results and perform a check:
| Mark source (score_source) | Description |
|---|---|
auto | Automatic LLM scoring result, default value |
manual | Scores manually reviewed by the instructor overwrite automatic grading |
During Instructor review, you can:
score_source is automatically marked as manualIf using grading data for research analysis, it is recommended to distinguish score_source as auto or manual to clarify the impact of grading source on the data.
The grading results for the AI interactive assignment are stored in the ai_interact_evaluations table. The main fields include:
| Field | Description |
|---|---|
criteria_scores | JSON format, scores and comments by dimension |
total_score | Total Score |
overall_comment | Overall Feedback |
score_source | auto (automatic) or manual (manual review) |
Conversation records and scoring results are fully retained, supporting subsequent learning analysis and research export.
A Student's AI interaction assignment grade is automatically assessed by the AI Interact module on the Uedu platform. The Instructor pre-sets the assignment mode (mode_code) and System Prompt. After the Student completes the assignment through multi-turn dialogue with the AI TA, the system uses an LLM (OpenAI gpt-5-mini) to grade each dimension according to the structured scoring criteria (scoring_criteria JSON), and outputs a JSON result containing the scores for each criterion (criteria_scores), the total score (total_score), and the overall comment (overall_comment), using response_format=json_object to ensure a consistent format. The scoring criteria may use the platform default or be custom-set by the Instructor. The Instructor can review the automatic grading result and manually verify it (score_source marked as auto or manual). See https://uedu.tw/doc/ai-interact for a detailed methodological explanation.
It is recommended to provide the following: