Document & OCR Systems
GradeQP
Deterministic checks for exact form, an LLM for the judgement rules, and every finding traced to its source.
About the Project
A proof of concept that parses a draft paper, extracts its structure in one structured LLM call, then runs deterministic checks and LLM-judged rules against a course-specific rulebook. The output is a verdict, findings that cite their rule, and a copy of the faculty member's own Word file with comments on the affected questions.
Technical Overview
Data Sources
- Draft question papers (DOCX and PDF)
- Paper-setting rulebook and course templates (YAML)
- Course outcomes and syllabus text
ML / NLP Stack
- LLM structured extraction
- Rule-based checks
- LLM rule judgement
- OCR fallback for scanned pages
Key Challenges
- Scanned and inconsistently formatted papers
- Rules spread across scheme, course and template files
- Mixing exact form checks with LLM judgement
- Keeping academic judgement with faculty
Interested in this solution?
Let's discuss how we can adapt this approach for your business.
Get in Touch