Scholarlink

The increasing demand for high-quality engineering education in India highlights the need for innovative teaching methods to bridge the gap between infrastructure and teaching effectiveness. Although active learning strategies have proven effective in STEM education, their adoption remains limited due to barriers such as time constraints and a lack of tailored resources. We introduce TeachBox, an always-on Retrieval-Augmented-Generation (RAG) chatbot that operationalizes discipline-specific pedagogy as an AI service. TeachBox embeds educator queries with OllamaEmbeddings, performs semantic search over a curated corpus of peer-reviewed studies and best-practice reports on STEM instruction, and ranks matches via a Facebook’s AI Similarity Search (FAISS) vector index. A three-agent pipeline (i.e., Subject, Scholar, and Summary) automatically scopes the topic, pulls the most relevant information, and compresses them into knowledge snippets that guide a ChatGroq mixtral-8 \(\times \) 7b-32768 language model to draft course-level plans, class-session blueprints, and evidence-linked active learning strategies. By surfacing sourced, context-aware pedagogy on demand, TeachBox lowers the cognitive and logistical overhead that currently limits deployment of active learning envisioned in India’s NEP-2020. The architecture is model-agnostic, interactive, and adaptive, positioning TeachBox as a scalable catalyst for evidence-based engineering education.

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

AI at the Chalkface: A Multi-agent RAG Assistant for Active Learning in STEM

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Rucha Joshi,
  • Suhani Shrivastava,
  • Nikita Thomas,
  • Malhaar Arora,
  • Ankur Nahar

摘要

The increasing demand for high-quality engineering education in India highlights the need for innovative teaching methods to bridge the gap between infrastructure and teaching effectiveness. Although active learning strategies have proven effective in STEM education, their adoption remains limited due to barriers such as time constraints and a lack of tailored resources. We introduce TeachBox, an always-on Retrieval-Augmented-Generation (RAG) chatbot that operationalizes discipline-specific pedagogy as an AI service. TeachBox embeds educator queries with OllamaEmbeddings, performs semantic search over a curated corpus of peer-reviewed studies and best-practice reports on STEM instruction, and ranks matches via a Facebook’s AI Similarity Search (FAISS) vector index. A three-agent pipeline (i.e., Subject, Scholar, and Summary) automatically scopes the topic, pulls the most relevant information, and compresses them into knowledge snippets that guide a ChatGroq mixtral-8 \(\times \) 7b-32768 language model to draft course-level plans, class-session blueprints, and evidence-linked active learning strategies. By surfacing sourced, context-aware pedagogy on demand, TeachBox lowers the cognitive and logistical overhead that currently limits deployment of active learning envisioned in India’s NEP-2020. The architecture is model-agnostic, interactive, and adaptive, positioning TeachBox as a scalable catalyst for evidence-based engineering education.