MBZUAI builds AI to save researchers drowning in papers
PaperCircle cuts through academic noise and helps put researchers back in control
#UAE #research - Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) has developed an open-source multi-agent AI system built to help researchers find, assess, organise and understand academic literature faster. ‘PaperCircle’ combines a discovery pipeline with a knowledge graph analysis tool to help researchers sift through thousands of resarch papers. MBZUA’s PaperCircle research paper has been accepted to ACL 2026, one of the most prestigious computational linguistics conferences, and nominated for an oral presentation.
SO WHAT? — The volume of scientific publishing has reached a scale that makes manual literature review genuinely unworkable. Millions of papers are now published annually across (science, technology, engineering, mathematics, and medicine (STEMM) fields, with growth accelerating sharply since 2004. Internationally co-authored papers alone jumped from 7,000 in 1980 to 440,000 by 2010. So, no researcher can reliably keep pace unaided. PaperCircle is a direct response to this growing challenge. Releasing the AI code as open source could put the solution in the hands of every research team on the planet, immediately.
Here are some key facts about PaperCircle:
Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) has released PaperCircle, an AI system designed to reduce the time and effort required to discover, evaluate and synthesise academic literature at scale. The system has been reeleased as open-source software, making it freely available to researchers worldwide.
PaperCircle runs on a multi-agent large language model (LLM) framework, where specialised AI agents handle different tasks working in a coordinated pipeline, from retrieval and scoring to knowledge graph construction and paper critique.
The discovery pipeline pulls from both offline and online sources, applying multi-criteria scoring and diversity-aware ranking to surface the most relevant research. Every step produces structured, reproducible outputs in JSON, CSV, BibTeX, Markdown and HTML formats.
PaperCircle’s analysis pipeline converts individual papers into structured knowledge graphs, with typed nodes covering concepts, methods, experiments and figures. This allows researchers to ask complex questions across an entire reading list, not just a single paper.
Meanwhile, a team of specialised review agents generates detailed critiques of papers, flagging strengths and weaknesses to help researchers prioritise their reading.
The PaperCircle research team’s paper has been accepted to ACL 2026 (the 64th Annual Meeting of the Association for Computational Linguistics), which will take place in San Diego from 2–7 July. The paper has been nominated for an oral presentation, a distinction awarded to a small share of accepted papers.
Benchmarking results show consistent performance gains as stronger agent models are used, with the system evaluated on paper retrieval and review generation using hit rate, MRR and Recall@K metrics.
The system has a known limitation in its review scoring: alignment with human reviewer judgements remains low, with correlation scores below 0.25 across models. However, MBZUAI is transparent about this, noting in its paper that larger models could help address the gap.
The research team include Komal Kumar, Hisham Cholakkal, Aman Chadha, Salman Khan, and Fahad Khan.
[Written and edited with the assistance of AI]
LINKS
PaperCircle (website)
PaperCircle research paper (arXiv)
PaperCircle code (GitHub)
PaperCircle data (Hugging Face)
Read more about recent MBZUAI research:
UAE lab breaks the speed barrier in AI video generation (Middle East AI News)
MBZUAI advances medical AI, cut training costs (Middle East AI News)
MBZUAI officially launches K2 Think V2 with mobile apps (Middle East AI News)
MBZUAI, Inception launch enhanced Nanda Hindi LLM (Middle East AI News)
MBZUAI releases K2 V2 open-source reasoning model (Middle East AI News)