RAG
Kitab Bot
Textbook answers with the page number, for students on every major syllabus.
Overview
Students use it before an exam for definitions, laws, and numericals, on a phone, without an account. When the evidence is weak, the assistant says the topic is not covered instead of guessing.
What we built
Kitab Bot is a production study assistant for students on the Federal, Punjab, Edexcel, and Cambridge syllabi. A student selects the board, class, and subject, asks a question from that textbook, and receives an answer drawn from the book, with the page number. The same page can be opened and checked.
Key features
- Board, class, and subject selection across Federal, Punjab, Edexcel, and Cambridge
- Answers drawn from the textbook, with a checkable page number
- Hybrid semantic and keyword retrieval with a reranker
- Says so when a topic is not covered, rather than guessing
- Cache for repeated questions on the same book
- Phone-first, no account needed
Architecture
The system indexes 51 official textbooks, about 13,600 pages. Each page is stored as a semantic embedding and a keyword index. A question is searched in both, limited to the selected book. The two result lists are merged, and a reranker keeps only the pages that match strongly. The answer is written solely from those pages. Repeated questions on the same book are served from a local cache when the wording is effectively the same.
Results
- 51 official textbooks indexed, about 13,600 pages
- On questions within the selected textbook, the cited page is correct about 98% of the time
Need something like this?
Tell us about the product or system you want to build.