RAG Enrollment Assistant (MESRS)

A national-scale, multilingual RAG assistant for Algeria's Ministry of Higher Education (MESRS) that lets new baccalaureate holders navigate university enrollment by chatting with the official circulaire and getting cited answers.

Project Overview

It runs a multi-agent RAG pipeline over the ministry's official documents, using multilingual embeddings for cross-language matching, re-ranking, and a confidence threshold that cites its sources or declines rather than guessing. An OCR stage makes the Arabic circulaire searchable, students can upload their BAC transcript in chat, and it is deployed on Dockerized, CI/CD-driven infrastructure. Retrieval was tuned to 99.5% hit@6 with 100% faithfulness and refusal accuracy on a 539-question set.

Key Features

Multi-agent RAG grounded strictly in the ministry's official knowledge base, with citations on every answer
Cross-lingual retrieval across Arabic, French, and English using multilingual embeddings, so a question in one language matches answers written in another
Hybrid retrieval: hand-curated bilingual summaries answer common questions first, with OCR'd circulaire passages as a fallback tier
Confidence threshold that steps back instead of guessing on off-topic or unanswerable questions
OCR ingestion for the non-machine-readable Arabic circulaire, plus in-chat BAC transcript upload (Ghostscript PDF to PNG, read by a vision model)
Instant eligibility checker (filière, wilaya, moyenne) across 100+ universities and 40+ programs
Tuned to 99.5% hit@6 with 100% faithfulness and 100% refusal accuracy on a 539-question benchmark