BanglaRAG is a multilingual retrieval-augmented generation system that is presented to learning management systems in universities in Bangladesh. This architecture is based on a combination of BanglaBERT and English embeddings along with page-level citation retrieval and deterministic citation mapping, and supports text and voice queries in Bangla and English. Caching, context truncation, and translation skipping in the system optimize the system, utilizing an 83.2% latency reduction and an average response time of 6.06 seconds. Testing of 230 test cases of textbooks in text shows a success rate of 82.61%, with Bangla queries being the highest of 90.91 and English queries being the best at 75%, yet citation rates are 100% reliable. Voice integration offers greater accessibility as it helps in meeting academic queries with good accuracy and efficiency. The findings suggest that BanglaRAG is to the authors' knowledge, among the first low-cost, domain-adaptive bilingual RAG system, offering practically deployment-ready support that can be trusted and is backed by evidence-based claims to serve students and educators.