Marcel is a lightweight, open-source conversational agent designed to support prospective students with admission-related inquiries. It aims to provide fast and personalized responses while reducing the workload of university support staff. The system is engineered for easy deployment in resource-constrained academic settings.

More information: Paper | Demo video

This repository provides the application related code. For experiments related to the EMNLP paper, please refer to the marcel-emnlp-experiments repository.

git clone git@github.com:aix-group/marcel-chat.git

Create a .env file based on the template and set the required variables.

cp .env.example .env
vi .env # Please refer to the .env file for configuration options

Simplified data is available in the data folder

The knowledge base defines the set of RAG sources used for your task. It must be a strictly valid and well-formatted .jsonl file. Each entry is a JSON object and must contain the following keys: url, content, favicon and

og: Metadata extracted via the Open Graph Protocol. This field is a dictionary and typically includes following keys:

• og:title: Title of the page
• og:description: Short description of the page
• og:type: Type of the object (e.g., "website", "article")
• og:url: Canonical URL (should match or normalize url)
• og:image: Image URL used for preview
• og:image:type: MIME type of the preview image

og can also be an empty dictionary.

{
    "url": "https://www.uni-marburg.de/en/studying/after-your-first-degree/masters-programs/application-for-a-masters-programme/master",
    "content": "# Master\n\n  \n\n\n\n\n",
    "og": {
        "og:site_name": "Philipps-Universit\u00e4t Marburg",
        "og:title": "Master",
        "og:type": "website",
        "og:description": "",
        "og:url": "https://www.uni-marburg.de/en/studying/after-your-first-degree/masters-programs/application-for-a-masters-programme/master",
        "og:image": "https://www.uni-marburg.de/@@site-logo/favicon.ico",
        "og:image:type": "image/x-icon"
    },
    "favicon": "https://www.uni-marburg.de/++theme++plonetheme.unimr/layout/favicon.ico"
}

If you like to use an FAQ retriever, you can build the FAQ. It must be a well-formatted .json file. Each object represents a single entry and must include the following keys: id, question and sources

[
  {
        "id": "faq-0001",
        "question": "Does the Master Data Science have a minimum grade requirement?",
        "sources": [
            "https://www.uni-marburg.de/en/studying/after-your-first-degree/masters-programs/degree-programs/datasciencems",
            "https://www.uni-marburg.de/en/studying/after-your-first-degree/masters-programs/application-for-a-masters-programme/eap/eap-data-science"
        ]
    }
]

Plese refer to backend instruction for admins configuration

[
    {
        "username": "admin",
        "hashed_password": "$argon2id$v=19$m=65536,t=3,p=4$FRnc99KQkJkRK+1RkTueyg$WbJ64c5iyXgHDy3nJyGqpEAyc1VsM8Q63ocF4XMRTF8"
    }
]

The application supports HTTPs-encrypted traffic. For development purposes, you can generate certificates like this:

mkdir certs
cd certs
openssl req -x509 -nodes -days 365 \
  -newkey rsa:2048 \
  -keyout selfsigned.key \
  -out  selfsigned.crt \
  -subj "/CN=localhost"
sudo chmod +r selfsigned.crt
sudo chmod +r selfsigned.key

docker compose --env-file .env up

The app will be available at: https://localhost:8080/

Refer to the following documents:

• Backend
• Frontend
• Scraper
• Docker Deployment: Docker

1. Welcome and Consent: Displayed on first use to inform and obtain user agreement.

2. Source Attribution: For answerable queries, Marcel displays the list of source documents used.

3. Feedback Mechanism: Users can rate individual responses (thumbs up/down). After three interactions, they can rate the overall conversation.

4. Answer Limitation: Marcel abstains from answering questions not supported by the underlying knowledge base.

Top: A dashboard shows key usage statistics over a customizable time period.
Bottom: List of conversations with links for manual review.

If you found any of these resources useful in your work, please consider citing our paper:

@inproceedings{trienes-etal-2025-marcel,
    title = "Marcel: A Lightweight and Open-Source Conversational Agent for University Student Support",
    author = {Trienes, Jan  and
      Derzhanskaia, Anastasiia  and
      Schwarzkopf, Roland  and
      M{\"u}hling, Markus  and
      Schl{\"o}tterer, J{\"o}rg  and
      Seifert, Christin},
    editor = {Habernal, Ivan  and
      Schulam, Peter  and
      Tiedemann, J{\"o}rg},
    booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
    month = nov,
    year = "2025",
    address = "Suzhou, China",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.emnlp-demos.13/",
    doi = "10.18653/v1/2025.emnlp-demos.13",
    pages = "181--195",
    ISBN = "979-8-89176-334-0",
    abstract = "We present Marcel, a lightweight and open-source conversational agent designed to support prospective students with admission-related inquiries. The system aims to provide fast and personalized responses, while reducing workload of university staff. We employ retrieval-augmented generation to ground answers in university resources and to provide users with verifiable, contextually relevant information. We introduce a Frequently Asked Question (FAQ) retriever that maps user questions to knowledge-base entries, which allows administrators to steer retrieval, and improves over standard dense/hybrid retrieval strategies. The system is engineered for easy deployment in resource-constrained academic settings. We detail the system architecture, provide a technical evaluation of its components, and report insights from a real-world deployment."
}