GETReason is a novel framework that goes beyond surface-level image descriptions to infer deeper contextual meaning from publicly significant event images. Our approach uses a hierarchical multi-agent reasoning system to extract geospatial, temporal, and event-specific information, enabling comprehensive understanding of visual narratives.

TRACEBACK: multi-agent attribution framework, traces table-based answers back to their supporting cells, providing fine-grained attribution at the cell level

(Findings of ACL 2025) TabXEval: an exhaustive, explainable rubric + two-phase framework (TabAlign → TabCompare) for table evaluation with TabXBench.

(WACV 2026) The Perceptual Observatory: Characterizing robustness and grounding in multimodal LLMs (MLLMs)

TabReX: Referenceless, explainable evaluation for LLM-generated tables via graph alignment + property-driven scoring (with TabReX-Bench)

(IJCNLP-AACL 2025) MAMMQA: Rethinking Information Synthesis in Multimodal Question Answering from a Multi-Agent Perspective

Please fill

REAR is a fast, LLM-free framework for multi-table retrieval that separates semantic relevance from structural joinability. By retrieving relevant tables, expanding with joinable ones, and refining noisy candidates, it consistently improves multi-table QA and Text-to-SQL performance—matching LLM-based methods at much lower cost and latency.