Academic RAG Platform
A hallucination-resistant RAG platform that serves students with verified, citation-backed answers. Built on a 3-agent pipeline that retrieves, re-ranks, and validates every response.
Students increasingly rely on AI tools for research, but generic LLMs hallucinate facts, misstate sources, and invent citations. In academic disciplines where accuracy is paramount, this is not acceptable.
Existing solutions either provide unverified answers or are too rigid to handle nuanced reasoning across domains. The gap between AI capability and academic reliability remains unbridged.
Embeds queries using FastEmbed (BAAI/bge-large-en-v1.5) and retrieves top-80 chunks from Qdrant vector store. Supports page-level precision and filename boosting.
Uses cross-encoder (Xenova/ms-marco-MiniLM-L-6-v2) to re-rank retrieved chunks to the top-20 most relevant passages. Eliminates semantic noise.
Adversarial validation agent confirms citations exist, claims are grounded, and answers engage with specific facts. Rejects hallucinated drafts and triggers rewrite.
Every response is challenged by a dedicated validation agent that confirms citations, verifies claims, and rejects hallucinated content before delivery.
Sources are cross-referenced with the knowledge base. Students receive verified, page-level citations — never fabricated references.
Answers follow Issue-Rule-Application-Conclusion structure tailored for legal education, with syllabus-aligned reasoning.
Dedicated environments per university with isolated document stores, chat histories, and module configurations.
Tailored to individual course modules — from Banking & FinTech Law to AI Governance — with curriculum-aligned content ingestion.
Full conversation persistence via MongoDB so students can return to previous queries and build on past answers.
Placeholder screenshots showing the platform interface. Actual product screenshots coming soon.
IRAC-formatted legal answers with verified citations and source references
Multi-tenant module selection with Banking & FinTech, M&A, AI Governance, and more
Real-time visibility into the 3-agent pipeline: Retrieval → Re-rank → Validation status
Page-level citations cross-referenced with uploaded academic materials
NyvLearn is deployed at QMUL across four academic modules, providing students with verified, citation-backed research assistance. The platform handles hundreds of queries per week with zero tolerance for hallucination.
We offer live demos for academic institutions, research collaborations, and technical partnerships. Let us show you how NyvLearn makes AI reliable.