# AutogradeX > AutogradeX is an AI-orchestrated digital evaluation system for universities and examination boards. A vision model transcribes handwritten answer books, an ensemble of independent grader models proposes marks against your rubric, and human evaluators finalise every mark. AutogradeX is a digital evaluation platform for university examinations. It manages the whole on-screen marking workflow, from receiving sealed answer-book bundles and scanning them to marking, moderation and result export. Its AI layer orchestrates several models instead of relying on one: a vision model reads the handwriting, multiple independent grader models mark each answer against the rubric, and an aggregator reconciles them into a proposal with an agreement score. Human evaluators review the scanned pages and decide every final mark. ## Multi-model AI orchestration 1. **Publish the marking scheme**: Coordinators enter each question's maximum marks, question text, model answer and rubric. Every model and every evaluator marks against the same scheme. 2. **Vision model reads the handwriting**: After a scan passes quality control, a vision model transcribes every page and records which question each page answers and how legible it is. 3. **Independent graders mark in parallel**: Several grader models, each a separate deployment you choose, mark the transcript independently against the rubric. None of them sees another's answer. 4. **The aggregator reconciles**: The orchestrator takes the median mark per question, measures how far the models agree, and lowers confidence wherever they disagree. 5. **The evaluator decides**: Evaluators see each model's mark, the agreement score and the cited pages beside the scan, then accept or change the proposal. Only the human-submitted mark counts. ## Digital evaluation workflow - **Hub receiving**: Sealed bundles are received with their seal number and manifest count, then counted. Any discrepancy is flagged. - **Chain of custody**: Every handover is recorded in a hash-chained ledger whose integrity can be verified with one click. - **Scanning & QC**: Books are registered by barcode. Multi-page PDFs or images are uploaded, then approved, or rejected for rescanning. - **AI orchestration**: A vision model and an ensemble of graders pre-mark every approved book. Failed runs retry automatically. - **On-screen marking**: Evaluators work with a real page viewer and annotations, and must view every page before submitting. - **Moderation**: Cases open on evaluator variance, random sampling, or human-vs-AI disagreement. - **Results & ERP**: Results export as CSV, Excel or PDF, or as checksummed ERP packages with reconciliation. ## Frequently asked questions ### What is AutogradeX? AutogradeX is a digital evaluation platform for university examinations. It manages the whole on-screen marking workflow, from receiving sealed answer-book bundles and scanning them to marking, moderation and result export. Its AI layer orchestrates several models instead of relying on one: a vision model reads the handwriting, multiple independent grader models mark each answer against the rubric, and an aggregator reconciles them into a proposal with an agreement score. Human evaluators review the scanned pages and decide every final mark. ### How does AutogradeX's multi-model AI orchestration work? Each answer book passes through a pipeline of models rather than a single one. A vision model transcribes the handwritten pages. Two or more independent grader models then mark the transcript against the rubric in parallel. An aggregator combines their marks per question using the median, calculates an agreement score, and flags questions where the models disagree. The result goes to a human evaluator as a proposal. ### Why use several AI models instead of one? A single model can be confidently wrong. When independent graders agree, the proposal is more reliable. When they disagree, AutogradeX lowers the confidence and tells the evaluator to look closely. Disagreement is treated as a signal instead of being hidden, and no single model decides a mark. ### Does AI decide students' marks in AutogradeX? No. AI only proposes marks, with confidence, rationale and evidence pages. An evaluator must view every page of the script and submits the final marks. If the final total differs too much from the AI proposal, a moderation case opens automatically for a senior moderator. ### Which AI models does AutogradeX use, and where does student data go? AutogradeX calls the Azure OpenAI deployments you configure in your own Azure subscription: one vision deployment for transcription and two or more grader deployments, which can be different models. Scans are stored in your private S3-compatible bucket, and marks, transcripts and audit records in your PostgreSQL database. ### What happens with poor handwriting or a bad scan? Scanning quality control rejects unusable scans for rescanning before marking starts. During transcription each page is rated clear, partial or illegible, and proposals that depend on hard-to-read pages get low confidence so evaluators check them carefully. ### Can a student trick the AI by writing instructions in an answer? No. Student writing is treated as untrusted data, and the graders are instructed to ignore any instructions or claims about marks inside answers. Every proposal is capped at the question maximum by the marking scheme, cross-checked by the other graders, and reviewed by a human. ### Can AutogradeX be used without AI? Yes. AI is switched on per subject and only when Azure OpenAI is configured. Without it, the full digital evaluation workflow works the same way with human marking. ### What does the AutogradeX digital evaluation workflow cover? Receiving sealed bundles at the evaluation hub with seal and count checks, a hash-chained chain-of-custody log, barcode registration, scanning and quality control, AI pre-marking, on-screen marking with annotations, single or double evaluation, moderation, and result reports with ERP export. ### How is AutogradeX deployed? As a single container with PostgreSQL and any S3-compatible storage. Run it on campus servers with Docker Compose or on Azure, AWS or Google Cloud container services. Database migrations run automatically, or you can apply the schema by hand. ## Links - [Home](https://autogradex.com/): product overview - [Privacy Policy](https://autogradex.com/privacy) - [Terms of Service](https://autogradex.com/terms)