Keywords: RAG, Agentic Systems, Large Language Models (LLMs), IISc M.Tech Online, LangChain, Student Assistant
Abstract: In the M.Tech Online program at the Indian Institute of Science (IISc), remote students face challenges in staying informed about academic updates and clarifying academic and administrative queries due to fragmented information across Microsoft Teams, institution website, intranet, handbooks and emails. Crucial information such as project deadlines, exam schedules, announcements and course-specific notifications is often buried in lengthy threads and dispersed, impacting student engagement and academic efficiency, particularly for new joiners who require timely clarification on academic and institutional procedures.
To address this problem, we propose the development of an agentic university coordination assistant, a conversational AI system that acts as a proxy for the current coordination team (IKEN). This assistant empowers students to ask natural language questions and receive accurate, context-aware, and up-to-date responses on both academic logistics and program-specific inquiries.
The system architecture integrates a multi-source ingestion pipeline that extracts, preprocesses and embeds heterogeneous data from institution website, Teams channels, handbooks, emails and intranet content. The data is transformed into vector embeddings using transformer-based models and stored in a vector database to support efficient semantic retrieval. A retrieval-augmented generation (RAG) pipeline, orchestrated using LangChain, dynamically constructs responses by interfacing with large language models (LLMs).
Evaluation of system output will be done using semantic similarity metrics to benchmark response quality across different model configurations and retrieval strategies.
This project aims to deliver an intelligent academic assistant that reduces operational overhead for the coordination team (IKEN), improves access to program-related information and enhances the remote learning experience for M.Tech Online students.
Submission Number: 10
Loading