AI pipeline for question banks and past papers

How we turned 235 exam papers into a searchable question bank with answers and solutions – using C# workers, LLM review and a self-healing job system.

235papers processed
14,000+questions structured
13,000+solutions written
24×7parallel workers

The problem

Past exam papers exist mostly as PDFs. Turning one paper into clean, structured questions – with options, diagrams, answer key, a worked solution and the right chapter and topic – takes a person many hours. Doing it for hundreds of papers by hand was not practical, and plain OCR or a single LLM call made too many silent mistakes.

What we built

  • Paper builder: splits each paper into questions, keeps diagrams and formulas, and reads the official answer key.
  • Solution writer: drafts a step-by-step solution and checks that its final answer matches the key.
  • Syllabus mapping: tags every question to its NCERT chapter and topic, so it appears on the right study page.
  • Independent review: a second model reviews each item; anything doubtful goes to a repair round or to a human.
  • Job system: SQL Server queues, parallel lanes on several machines, and a watchdog that restarts stuck jobs from the last good step.

Technology

  • C# .NET workers and an Avalonia desktop console for monitoring.
  • SQL Server job tables with clear status for every paper and question.
  • Several LLMs for drafting and review, with locked prompts and source text so answers stay grounded.
  • Laravel import commands that publish approved items to the student website.

Result

The pipeline has processed 235 JEE papers into more than 14,000 questions, now live on JEE NEET AI. The same approach works for any question bank: school boards, competitive exams, corporate training or certification tests. Read more about our EdTech content automation and AI automation & LLM agents services.

Have something to build or automate?

Tell us about it – we reply within one working day.

Start a conversation