DeepTutor Online Tutorial: HKU’s Open‑Source Multi‑Agent Interactive Learning Assistant

Problem

Learners often have to search dense textbooks and large numbers of papers to locate key knowledge, lack intuitive explanations for complex concepts, and must switch among multiple tools when conducting research.

Technical approach

DeepTutor is an open‑source personal learning assistant that combines a multi‑agent architecture with several toolchains:

Retrieval‑augmented generation (RAG) for grounding large‑model responses in external documents.

Real‑time web search to obtain up‑to‑date information.

Academic paper database access for scholarly retrieval.

Users submit natural‑language requests (e.g., solving a problem, planning a study path, generating exercises, drafting a research report). The system automatically performs intent parsing, retrieves relevant information from the chosen toolchain, and produces a structured output. This task‑centric interaction replaces the traditional function‑centric workflow of existing e‑learning tools.

Core functionalities

Massive document Q&A : upload textbooks, papers, or technical documents to build an AI knowledge base; multi‑agent collaboration provides answers with precise citations.

Interactive learning visualization : convert complex concepts into visual tools that support personalized Q&A and context‑aware dialogue.

Knowledge reinforcement and exercise generation : generate targeted quizzes and practice questions that match the learner’s proficiency and simulate real‑exam styles.

Deep research and creative generation : use RAG, web, and paper retrieval to explore topics in depth, identify knowledge gaps, and suggest potential research directions.

Deployment steps (HyperAI tutorial environment)

Open the HyperAI homepage, navigate to the “Tutorial” page, and select “DeepTutor Personal Learning Assistant”. Click “Run this tutorial”.

On the tutorial page, click the top‑right “Clone” button to copy the repository into your container.

Select the compute node “NVIDIA RTX 5090‑4” and the “vLLM” image, then click “Continue job execution”.

Wait for the job status to become “Running”, then click “Open Workspace” to launch the Jupyter workspace.

In the workspace, open the README file and click “Run”. The demo executes and exposes an API endpoint for interactive use.

Access links

Demo URL: https://go.hyper.ai/sC4nC Open‑source repository:

https://github.com/HKUDS/DeepTutor