AI Descriptive Answer Evaluation

Modified on Mon, 8 May, 2023 at 12:06 PM

How to use AI Descriptive Answer Evaluation

AI Descriptive Evaluation automates the grading of descriptive (typed) answers in online exams. The AI compares each student response to a model answer and awards marks for accuracy, completeness, keyword coverage, and structural quality. Faculty review and override before finalizing.

How it works

Manual seed evaluation — faculty manually evaluate the first 20–25% of answer sheets so the AI can learn the marking patterns for this specific exam.
AI learns the pattern — the system analyses your manual scores against the model answer and learns your scoring style.
Auto evaluation — the AI scores the remaining 75–80% of answer sheets using the learned pattern.
Quality assurance — random sampling and moderation ensure the AI scores stay aligned with examiner expectations.

Step 1 — Configure the exam

Log in as admin and create or open the descriptive exam.
Add descriptive questions, each with a clear model answer and the maximum marks.
Open Exam Settings → AI Evaluation and toggle Enable AI Descriptive Evaluation.
Choose evaluation strictness: Easy / Default / Moderate / Strict. Default is fine for most exams.
Set the manual seed percentage — default 25%. The first N% of answers will be flagged for manual evaluation before AI takes over.
Save.

Step 2 — Manual seed evaluation

After the exam ends, log in as examiner.
Open the assigned answer sheets — the first N% are flagged "Manual seed required".
Read each answer, award marks question-by-question, and submit.
Once the seed quota is reached, the system enables AI evaluation for the rest.

Step 3 — AI evaluation runs

The system processes remaining answers in the background.
Progress is visible in Evaluation Dashboard.
Each AI-scored answer is queued for human review.

Step 4 — Human review and override

Faculty opens the AI-scored queue.
For each answer, the dashboard shows: the candidate's response, the AI's score per question, and the AI's reasoning.
Click Accept to keep the AI score, or Override to enter your own.
Optionally add feedback comments the candidate will see.
Submit to finalize.

What the AI evaluates

Dimension	What it checks
Accuracy	How well the answer matches the model answer's facts and claims.
Keyword detection	Whether key concepts and terminology appear in the answer.
Structure	Logical flow, paragraphing, introduction-body-conclusion shape.
Semantic relevance	Does the answer mean the right thing, not just contain the right words.
Partial credit	Awards proportional marks where the answer is partly correct.
Plagiarism / copy detection	Flags answers that appear copied from another candidate or from external sources.

Multi-language support

The AI evaluates English, Hindi, Marathi, Tamil, Telugu, Bengali, Gujarati, and several non-Indian languages. The model answer language must match the language candidates write in.

Cheating detection

Built-in plagiarism detection flags answers that are too similar to other candidates' answers or to public sources. Flagged answers are surfaced in the review queue with the comparison.

Common questions

Can I run AI evaluation on past exams that have already been manually evaluated?
Not retroactively — the manual seed and AI evaluation are interleaved during the same evaluation cycle. For past exams, you can re-run as a re-evaluation exercise.

Does the AI evaluate handwritten descriptive answers (image uploads)?
For typed answers, yes — directly. For uploaded handwritten images, see OSM Auto Evaluation which adds OCR before scoring.

Can students see the AI's reasoning?
By default candidates see their final score and any examiner feedback. The AI's internal reasoning is admin-side only. You can opt to share AI feedback with students from Exam Settings → AI Feedback Visibility.