For edtech platforms & developers
Handwritten answer checking, as an API.
{
"submission_id": "5a7e2c90-1b4d-4e8f-a3c2-7d9e0f1a2b3c",
"status": "graded",
"needs_review": true,
"finalized": false,
"totals": { "awarded": 38, "max": 80, "percentage": 47.5 },
"questions": [
{
"label": "30",
"awarded": 0.5,
"max": 3,
"source": "ai",
"confidence": 0.88,
"needs_review": false,
"extracted_answer": "1. Temperature 2. Humidity 3. Precipitation 4. Atmospheric Pressure 5. Wind",
"feedback": "Only weather elements listed; no factors of climate explained.",
"criteria": [
{ "name": "Names valid factors of climate", "awarded": 0.5, "max": 1.5, "reason": "Lists weather elements, not factors." },
{ "name": "Explains how each factor affects climate", "awarded": 0, "max": 1, "reason": "No explanation." },
{ "name": "Uses a correct example", "awarded": 0, "max": 0.5, "reason": "No example." }
]
}
],
"checked_copy": { "available": true, "download_path": "/submissions/5a7e2c90-…/checked-copy" },
"credits_charged": 7
}The problem today
OCR alone fails on handwriting
Printed-text OCR cannot read student handwriting, and a grader fed garbage text invents answers.
A score is not a product
Learners want to see their own page marked up, not a JSON blob turned into a table.
Building it in-house is a project
Page reading, answer location, rubrics, consistency, annotation, retries and a review screen for teachers are each a project of their own.
How it fits your workflow
- 1.
Create the exam
POST /exams with the questions as JSON: max marks, answer keys for objective questions, and a rubric or model answer for each long answer.
- 2.
Upload each copy
One PDF per candidate, up to 50 MB, through presigned upload URLs (1 to 100 files per call). Typed tests send the answers as text instead.
- 3.
Submit and follow
Submit each copy against your own candidate id. You get 202 and a fixed-price quote at once; poll the submission, or the updated_since feed for many.
- 4.
Read, review, finalize
Question-wise marks with reasons, feedback, confidence, needs_review and the checked copy PDF. Teachers review in your app or the Vacademy dashboard; then you finalize.
What changes
- Question-wise marks, criteria with reasons, feedback and confidence in JSON
- The student's own copy checked in red pen, as a PDF to show learners
- A needs_review flag on the answers a teacher should look at
- Teacher review without building a screen: API exams appear in the Vacademy dashboard
- ₹1 / $0.01 per handwritten page, quoted before grading; failed and cancelled copies are free
Worked example
20,000 answer sheets a month, 5 pages each
20,000 copies × 5 pages × 1 credit
1,00,000 credits
Credits come with every plan · failed copies free
Work out your ownFrequently asked questions
How do I get API keys?+
We switch the Evaluation API on for your institute: write to [email protected] or request access on the demo page with your expected monthly volume. An institute admin then creates keys in the Vacademy dashboard (Settings → Integrations → API keys), each with only the scopes it needs. There is no sandbox: every key is live and every graded copy is billed.
Is it synchronous?+
No. A handwritten copy usually takes a few minutes, so grading is asynchronous: the submit call returns 202 straight away with the submission id, its place in the queue, an estimated ready time and the fixed price. Poll GET /submissions/{id}, or the GET /submissions?updated_since= feed when many copies are in flight. Webhooks are not available yet.
What do I get back?+
The status and totals, and for each question: marks awarded out of the maximum, each rubric criterion with its marks and reason, feedback, the answer as read, a confidence score and a needs_review flag. For handwritten copies, GET /submissions/{id}/checked-copy returns the red-pen PDF. Marks are drafts until you call finalize.
What can I send?+
Handwritten copies as PDFs only, one per candidate, up to 50 MB; phone photos must be combined into a PDF first, and the API does not yet split a bundled scan or match copies by name. Copies up to 40 pages are graded normally; 41 to 80 pages are graded and flagged for review. Typed tests send each answer as plain text. Questions go in as JSON, not as a question-paper PDF. English answers only for now.
How is the API priced?+
1 credit per page of a handwritten PDF, blank pages included, and 1 credit per non-blank typed long answer; objective answers are free. At the standard rate a handwritten page costs ₹1 in India (excluding GST) or $0.01 elsewhere, the same as in the dashboard. The price is fixed and quoted before grading. Failed and cancelled copies are not charged; re-evaluating a copy is charged again. Details at docs.evalezy.com/platform/pricing.
See your own copies checked.
Book a 20-minute demo. Bring a question paper and a few scanned answer sheets, and watch Evalezy check them in red pen.