Scholarlink

Retrieval-Augmented Generation (RAG) can substantially improve factuality and contextuality in educational question answering. We present an intelligent study assistant that integrates a Next.js front end, a FastAPI back end, and ChromaDB for vector retrieval with LLaMA 3 70B for response synthesis. Queries are embedded, matched via cosine similarity to course materials, and passed as grounded context for generation. The system also supports code generation and live preview for programming tasks. In our internal evaluations, the RAG pipeline reduced latency and improved alignment between generated answers and retrieved sources, with higher user-rated usefulness than a non-retrieval baseline. We detail architecture, retrieval and prompting strategy, guardrails, and deployment considerations, and we report a comparative summary of LLaMA 2 versus LLaMA 3 for this task.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Intelligent Study Assistant for Contextual Answer and Code Generation Using RAG Models and LLMs

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ishita Jain,
  • Harsh Jajal,
  • Shreedhar Joshi,
  • Nishit Shetty,
  • Manya Gidwani

摘要

Retrieval-Augmented Generation (RAG) can substantially improve factuality and contextuality in educational question answering. We present an intelligent study assistant that integrates a Next.js front end, a FastAPI back end, and ChromaDB for vector retrieval with LLaMA 3 70B for response synthesis. Queries are embedded, matched via cosine similarity to course materials, and passed as grounded context for generation. The system also supports code generation and live preview for programming tasks. In our internal evaluations, the RAG pipeline reduced latency and improved alignment between generated answers and retrieved sources, with higher user-rated usefulness than a non-retrieval baseline. We detail architecture, retrieval and prompting strategy, guardrails, and deployment considerations, and we report a comparative summary of LLaMA 2 versus LLaMA 3 for this task.