A scholar-driven AI notebook for assisted and transparent research

In December 2025 GoTriple.eu introduced an AI-powered chatbot for its users: it allows them to use Large Language Models (LLMs) to analyse and extract insights on collections of documents. These collections can be created by selecting from the over 20 million publications indexed in GoTriple. The tool is multilingual and uses the open-source LLM DeepSeek-R1 (685 billion parameters) made available through the GRAPHIA Horizon project, coordinated by OPERAS. Groundwork for the chatbot was initially done via the FASCA project also led by OPERAS. The tool is free to use for registered GoTriple users. Users can use it in their research workflows, from early exploration of scientific literature to the extraction of relevant knowledge via an AI-assisted analysis process.

The notebook becomes a work environment for conducting focused analyses on the documents: activities like systematic literature review, text translation, extraction of entities from the text, to be used in data analytics tasks, can be easily performed via a chatbot-like interface. The tool maintains the history of interactions, providing a familiar conversational interface.

The chatbot has been integrated into the GoTriple White Label solution, an open-source software that allows communities from any scientific domain to create their own discovery platform, reusing  GoTriple.eu. The White Label solution is one of the results of the Horizon LUMEN project, aiming to foster interdisciplinary research and innovation across the EOSC ecosystem.

In order to further expand the functionality of the tool, the LUMEN team is integrating external services through MCP (Model Context Protocol), an open protocol that standardizes how AI models interact with external tools and data sources, by exposing them as structured, machine-readable capabilities. The assumption here is that GoTriple will be able to benefit from the expanding MCP ecosystem by providing an MCP-ready solution, capable of leveraging servers and code written by other trusted organisations, such as other EOSC projects.

This approach makes the notebook extensible, allowing to offer an expanded portfolio of services: users can pull documents from multiple repositories (e.g. from Zenodo), retrieve external data (e.g. from national statistical services), produce diagrams, and also push AI-extracted structured outputs (e.g., JSON) to systems like GitHub, enabling reproducible workflows, collaboration, and automated pipelines. For example, a researcher can start from a curated collection of documents, extract scalar data embedded in them to generate charts for a publication. Or they could retrieve external data from national statistical services or parliaments, that connects directly with the papers in their collection.

In this poster we will present:

• the current design of the GoTriple chatbot, and outline its workflow in graphical form, including for the creation of collections

• the philosophy of adoption of MCP for the GoTriple chatbot

• a graphical workflow of how a user could leverage the chatbot and MCP to perform their analysis on literature collections

The poster will contribute to the current discussion about adoption of AI across the OPERAS services, and how current research projects can support the OPERAS community in the transparent and FAIR adoption of AI solutions.