Comprehensive technical guides to help you navigate handwriting extractions and semantic grading workloads.
To begin utilizing the EvalAI platform:
Every evaluation job requires a model key/question paper structure. Upload a PDF of the question paper detailing question indexes and maximum marks per question. Specify rubrics and sub-questions (e.g. Q1.a, Q1.b) clearly in the form settings.
Ensure scanned answer sheets comply with these standards to optimize OCR precision:
For class-wide evaluations, navigate to the upload section and choose the "Bulk ZIP" option. You can upload up to 500 sheets in a single job. The system processes pages concurrently, updating the queue log in real-time.
Prepaid token credits are consumed on a page-by-page basis: 1 token for character extraction (OCR) + 1 token for AI semantic comparison against your model answers. Total is 2 tokens per scan page.
Once files are uploaded, our spatial layout OCR analyzes margins, segmenting handwriting sections. It filters noise (like eraser marks or ink blotches) and translates handwriting into clean ASCII digital text transcripts.
EvalAI feeds digital text transcripts and question rubrics to our semantic model parser (such as Google Gemini). The model evaluates contextual correctness, awards partial credit, adds descriptive comments, and sums up scores.
Once grading completes, audit scores on the job dashboard. Perform manual marks overrides if needed. Click the "Export Excel" action button to download complete marksheets (XLSX) with student names, roll numbers, question scores, and totals.
For billing queries, subscription details, and team management resources, consult our Help Center portal.