Aceprep

Past paper content is spread across PDFs and sites. Students need fast help and feedback that matches official marking expectations, not generic tutoring disconnected from schemes.

Without structured metadata, search and tutoring cannot reliably tie explanations back to marks, parts, and examiner commentary.

I built a pipeline to collect and normalize papers, indexing by question structure: parts, marks, teacher commentary, and marking schemes for precise retrieval.

A Q&A layer retrieves relevant indexed segments and uses LLMs to answer questions or validate student work, grounded in schemes and commentary.

Dockerized ingestion and indexing pipeline with PostgreSQL for structured storage. Containerized deployment supports scaling the Q&A layer independently from ingestion.

Students interact with a corpus that behaves like a structured exam archive rather than unstructured PDFs.

Answers and validation reflect examiner language and scheme logic, improving trust and study quality.