Smart Academic and Placement Assistant using Graph-RAG

The Smart Academic & Placement Assistant was developed as a centralized academic resource and placement preparation platform integrating a React-based  user  interface,  a  TypeScript/Node.js backend, and a generative AI chatbot. The application provides role-based access for students and administrators, allowing administrators to publish academic PDF resources and students to access learning materials and ask questions through the AI Academic Assistant. The chatbot uses the Google Gemini API through the Google Gen AI SDK, with the configured model gemini-3.8-flash. The backend retrieves relevant passages from indexed PDF pages and supplies the retrieved context alongside the user’s query to the language model. The answer-generation pipeline is designed to produce a direct explanation, supporting details, and examples where relevant. Retrieved source references are displayed with the PDF filename, page number, and text excerpt, enabling students to verify the information against the uploaded academic notes. The PDF retrieval component uses the pdf-parse library for page-wise text extraction and sql.js for SQLite-based storage of indexed page content. The chatbot response is returned through the Express REST endpoint POST /api/chat. The frontend then presents the generated explanation and its associated PDF references in the chatbot interface.

[1]   P. Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” Advances in Neural Information Processing Systems, vol. 33, pp. 9459–9474, 2020. [2]  H. Han et al., “Retrieval-Augmented Generation with Graphs (Graph-RAG),” arXiv preprint arXiv:2501.00309, 2025. [3] Y. Wang et al., “Knowledge Graph-based Question Answering Systems,” IEEE Transactions on Knowledge and Data Engineering, vol. 34, no. 5, pp. 2103–2115, 2022. [4] T. Kipf and M. Welling, “Semi-Supervised Classification with Graph Convolutional Networks,” International Conference on Learning Representations (ICLR), 2017. [5]   A. Grover and J. Leskovec, “Node2Vec: Scalable Feature Learning for Networks,” Proceedings of the 22nd ACM SIGKDD Conference, 2016. [6]   B. Perozzi et al., “DeepWalk: Online Learning of Social Representations,” Proceedings of the 20th ACM SIGKDD Conference, 2014. [7]  Y. Yasunaga et al., “QA-GNN: Reasoning with Language Models and Knowledge Graphs,” NAACL, 2021. [8] X. Wu et al., “Graph-based Retrieval for Knowledge-Intensive Tasks,” IEEE Access, 2023.