Poster Session (Demos)

GutBrainKB is a web platform based on a structured KB to aid clinicians and researchers in studying gut¬Cbrain interactions. It features a 79K-triple KB from expert-annotated literature in the GutBrainIE dataset (BioASQ Lab @ CLEF 2025), offering a reliable resource for exploring the gut-brain axis. Users can explore through natural language and faceted search, investigating entity relationships, accessing scientific literature, and visualizing connections with interactive graphs. GutBrainKB combines a semantically validated KB with an accessible interface, providing an efficient tool for analyzing gut¬Cbrain interactions and their impact on neurological and psychiatric disorders.

PyTerrier provides a declarative framework for building and experimenting with information retrieval (IR) pipelines. In this demonstration, we highlight several recent higher-level pipeline operations that improve their ability to be programmatically inspected, visualized, and integrated with other tools (via the Model Context Protocol, MCP). These capabilities aim to make it easier for researchers, students, and AI agents to understand and use a wide array of IR pipelines.

Retrieval-augmented generation (RAG) systems promise factually grounded answers, yet evaluating their quality remains difficult. Automated metrics and LLM-as-judge approaches offer scalability but risk circularity, benchmark leakage, and loss of diversity. Human assessors, meanwhile, often struggle to notice subtle omissions or hallucinations when responses appear linguistically fluent and confident. We present Talmud-IR, a novel user interface inspired by the dialogic structure of the Talmud. It visualizes RAG outputs as a central text surrounded by layers of evidence, commentary, and meta-assessment, enabling sustained human--LLM discussion about system quality and failure priorities. The prototype supports comparative RAG evaluation, collaborative exploration of ``unknown unknowns,'' and pedagogical use for teaching critical reading of AI-generated content.

Recommender systems researchers rely heavily on general-purpose libraries that facilitate data preprocessing, model training, and evaluation. However, existing frameworks often suffer from fragmented data handling, inconsistent preprocessing, limited interoperability, and poor dataset referencing, which hinder reproducibility and comparability between studies. We present OmniRec, an open-source Python library designed to address these limitations. OmniRec provides standardized access to more than 230 datasets, a unified and flexible preprocessing pipeline, and seamless integration with multiple state-of-the-art recommender system frameworks, including RecPack, RecBole, Lenskit, and Elliot. Its modular architecture allows researchers to easily integrate new datasets, customize preprocessing steps, and external model interfaces. By combining ease of use, transparency, and reproducibility, OmniRec simplifies experimentation and fosters a more open and collaborative ecosystem for recommender systems research and practice.