Overview
Feedback Aide is an embeddable API for assessing a learner response against a rubric using Learnosity’s AI engine. Feedback Aide is a safe and efficient assistant for evaluating and providing quality feedback on essays.
Learnosity's AI engine generates an initial score for a learner response, given the question stem and a grading rubric, as well as actionable, written feedback. Learnosity's FeedbackAide API then helps graders easily edit the generated scores and feedback.
Screenshot 1: Feedback Aide showing the rubric / assessment criteria
How it works
Feedback Aide takes an essay 'stimulus', a scoring 'rubric' and a user 'response'. It then evaluates the 'response' to the 'stimulus' against the 'rubric' and identifies the appropriate rubric levels for the essay and provides constructive feedback based on the essay.
This can then be reviewed, and modified by a human grader prior to be saved.
Once the feedback has been finalized it can then be presented to the learner in a simple read-only interface.
Screenshot 2: Feedback Aide showing the automated grading decisions (in orange)
Screenshot 3: Feedback Aide showing the manual grading decisions (in blue)
Features
- AI enhanced feedback and grading of analytic and holistic rubrics with essays between 100 and 10,000 characters.
- Two different models depending on use case:
- standard-essay - flexible and efficient model with good accuracy and feedback over a wide range of essay types and rubrics.
- advanced-essay - our most advanced model with excellent accuracy and feedback over a wide range of essay types and rubrics.
- Topic Detection - identify essays that are not on topic and explains why it is considered off topic.
- Moderation & Safety - optional AI-driven moderation helps ensure that sensitive, inappropriate, or crisis content is flagged and managed efficiently. Content moderation with Feedback Aide
- Secure and private - responses are not used to train LLM foundation models.
- Designed for localisation and internationalisation: internationalization - Feedback Aide API
-
Full Grading and Review workflow:
-
Grading state. This is meant for the first scoring of a learner’s response. It initializes with a blank rubric and a score of –/max. If this session has previously been initialized it will restore any feedback that has previously been saved.
-
Learner review state. This is a read-only view of the populated rubric, overall score, and written feedback. In this state, the learner will not be able to see which scores and feedback were originally generated by AI and which were human-generated.
-
Screenshot 4: Feedback Aide in learner review state
Current Models
The following models are currently available.
| Model | Description |
| standard-essay | Flexible and efficient model with good accuracy and feedback over a wide range of essay types and rubrics - currently points to `standard-essay-2024-11-22` |
| standard-essay-moderation | Standard essay model with moderation enabled. Note this is replaced by using model options to enable moderation (deprecated) |
| standard-essay-2024-11-22 | The November release improves accuracy for most use cases, and also adds topic detection which flags essays that are completely off topic. |
| standard-essay-2024-08-05 | standard-essay currently maps to this and is recommended for all uses cases as performs better than previous model. |
| standard-essay-2024-06-30 | Legacy version of standard essay (deprecated) |
| advanced-essay | Our most advanced model with excellent accuracy and feedback over a wide range of essay types and rubrics - this will point to the best performing advanced model - currently mapped to `advanced-essay-2024-11-22` |
| advanced-essay-moderation | Advanced essay model with moderation enabled. Note this is replaced by using model options to enable moderation (deprecated). |
| advanced-essay-2024-11-22 | The November release improves accuracy for most use cases, and also adds topic detection which flags essays that are completely off topic. |
| advanced-essay-2024-08-05 | The August advanced-essay improves accuracy and is significantly better than previous model. |
| advanced-essay-2024-06-30 | Legacy version of advanced essay model. (deprecated) |
Preview Models
Note these models will be removed but are useful for early access to features
| Model | Description |
| preview-standard-essay-ISCED1-2024-12-16 | Preview of standard ISCED1 (primary/elementary) focussed model which shows improved accuracy for younger learners. |
| preview-advanced-essay-ISCED1-2024-12-16 | Preview of advanced ISCED1 (primary/elementary) focussed model which shows improved accuracy for younger learners. |
| preview-standard-essay-ISCED3-2024-12-16 | Preview of standard ISCED3 (high school) focussed model which shows improved accuracy for high school level learners. |
| preview-advanced-essay-ISCED3-2024-12-16 | Preview of advanced ISCED3 (high school) focussed model which shows improved accuracy for high school level learners. |
Troubleshooting
- Manual grading: If AI grading isn't enabled or is unavailable, a human grader can manually score the response using the rubric.
- Unsupported essay: If the response is less than 100 characters or over 10,000, Feedback Aide will not grade the essay. Graders will be shown a warning message, and can manually grade the response.
- Disruption: If there is an interruption in service from the AI engine, the grader will see an error message informing them of the disruption. They have the option to regenerate, resend the information to the engine, without using additional credits, or to manually grade.
Release Notes
The detailed release notes are available here: Feedback Aide Release Notes