AI grading for handwritten answer sheets, built for Indian classrooms.
Kwilo AI reads scanned or photographed answer sheets and returns marks, per-question feedback, and a rubric audit trail — without replacing the evaluator.
What the grader does
Upload a batch of scanned answer booklets in PDF or JPEG format. Kwilo's OCR pipeline reads handwritten English answers. Each answer is scored against your rubric, with highlighted deductions and a summary score per learner. Faculty review AI marks and approve or override, the final grade is always a human decision.
Built for exam coordinators, HoDs, and faculty
A 60-seat batch at an engineering college generates roughly 3,000 answer sheets per semester exam. Manually evaluating each one takes 4–6 weeks. Kwilo reduces that to under 72 hours. HoDs get department-level analytics — which questions every learner missed, which questions distinguish top quartile from the median. Exam cells get an audit trail for re-evaluation disputes.
Data sovereignty and DPDP compliance
Answer sheets contain learner PII. Kwilo stores uploaded sheets in your institution's GCP or Azure tenant — not on shared infrastructure. Sheet data is retained only for the duration you configure and deleted on schedule. All processing meets DPDP Act 2023 obligations for purpose limitation and storage restriction. No learner data leaves your designated region.
Early results from our pilot colleges
Kwilo is grading handwritten answer sheets for our first cohort of college partners, scoring against a rubric instead of a blanket mark so every deduction traces back to a clause. Early feedback from faculty: turnaround is dropping from weeks to days, and re-evaluation conversations get shorter because the reasoning is already on the page.
Request a demo
View pricing
Frequently asked questions
Kwilo currently grades handwritten answers written in English. The model is fine-tuned on examination handwriting, not printed text, so it handles rushed cursive and the mixed formatting common in science and engineering answers. Support for additional Indian scripts is on the roadmap.
Yes — overrides are the default workflow. AI marks are a first draft. Faculty review each answer, override marks where needed, and the final grade reflects the human decision. Every override is logged so your audit trail shows both the AI proposal and the faculty judgement.
PDF (multi-page), JPEG, and PNG. Resolution must be at least 150 DPI. Scans at 200–300 DPI produce the best OCR accuracy. If your scanning equipment produces lower-resolution output, Kwilo's upscaling step can compensate in most cases.
Rubrics are created in the platform by the faculty member who set the paper. Each question gets a maximum mark, a list of expected concepts, and optional partial-credit bands. Existing CBCS/VTU rubric formats can be imported from a spreadsheet template. No programming required. For descriptive questions, rubrics can include a model answer with a keyword and concept list. The grader checks for the presence and accuracy of those concepts, not just verbatim keyword matches, so partial marks are awarded fairly when learners phrase answers differently from the model. Faculty can preview how a rubric scores a sample answer before running it across the full batch, which catches calibration errors early.
Kwilo is used as a supplementary evaluation tool, not as a replacement for the legally mandated examiner. The examiner remains the authoritative grader; Kwilo speeds up their work and provides the audit trail. This is consistent with AICTE and UGC guidelines on AI-assisted evaluation.