Skip to main content

Lecture Linker (2024)

Besser Digital: Design and Development of a Full-Stack Application

A platform (LectureLinker) was developed that enables educators to create their own chatbots for students. Instructors upload their lecture recordings as videos and receive an individual link to share with their students. Through this link, students can activate the chatbot for the lecture. In addition to standard chatbot functions, it uses the lecture recordings as an information source and links to relevant sections. Students can use these links to verify relevant information or learn more. The chatbot uses generative artificial intelligence that runs entirely on a University of Siegen server in compliance with data protection regulations.

Funding & Resources​

I secured project funding from the University of Siegen's Besser Digital program. This project was developed in collaboration with the Chair of Operating Systems and Distributed Systems and Computer Science Education.

With the funding, I acquired a workstation equipped with two NVIDIA RTX 4090 GPUs. This machine runs the entire solution locally: the Large Language Models, the transcription model, the embedding models, and the full-stack web application. Additionally, I hired a student assistant who supported me during the development phase.

Access LectureLinker via the University of Siegen VPN at: https://lecturelinker.bshefl2.bs.informatik.uni-siegen.de

Technical Implementation​

System Architecture

LectureLinker runs entirely locally. We use the following models:

  • Large Language Model: qwen2.5:32b-instruct-q4_K_M
  • Embedding: bge-m3
  • Transcription: OpenAI Faster-Whisper

For the implementation of Retrieval Augmented Generation (RAG), we use the Contextual Retrieval technology introduced by Anthropic. Only reranking is not yet implemented.

For Instructors​

Course Overview for Instructors

Instructors should contact Sven Jacobs to be registered as an instructor. After that, you can create your own courses, upload videos, and release your chatbot.

For Students​

Chat Interface

Students can access the chatbot through a link provided by their instructor.

Evaluation​

Evaluation Results

Since informed consent from students was not yet available, we could not evaluate the chat histories. Instead, we systematically evaluated 120 synthetic question-answer pairs. For the future, an evaluation with questionnaires at the end of the lecture period with students is planned. We would be happy to cooperate with you.