Exeter to Help Lead £11.5m AI Project Delivering Living Evidence for Faster Policy Decisions

Exeter to Help Lead £11.5M AI Evidence Project

Exeter academics are helping lead a £11.5 million initiative to change how governments consume and act on research evidence. The Mobilising Evidence Through AI and User-informed Synthesis (METIUS) project targets time-critical decisions in climate, education, and public safety.

Led by Queen's University Belfast with the University of Exeter, and supported by NIHR Applied Research Collaboration South West Peninsula (PenARC), the project tackles a familiar problem: research volume outpaces the ability to find, assess, and apply it.

Teams at the Universities of Exeter and Newcastle will steer the Methods Work Package, developing approaches for "living evidence syntheses" that stay current and deliver timely, usable insights. Professor Ruth Garside said: "The sheer volume and range of new research is a real challenge for policymakers who need to act quickly on key issues like climate change and international development. We're hoping to develop concrete methods that will allow interaction between AI and researchers to help us cut through that noise and synthesise evidence more effectively - ultimately ensuring that critical decisions are informed by the most relevant and up-to-date scientific findings."

Professor Jo Thompson-Coon added: "We are excited that PenARC is part of this transformative effort. Evidence synthesis has always been at the heart of what we do - METIUS gives us the opportunity to scale it up, speed it up, and ensure evidence truly reaches decision-makers when and where it matters."

Why this matters for IT and development teams

METIUS isn't just policy talk-it's an engineering challenge. Making living syntheses work at scale demands solid data plumbing, reliable model orchestration, and repeatable evaluation.

• Ingestion at scale: full-text scraping, PDF parsing, deduplication, metadata normalization, and multilingual support.
• Search that goes beyond keywords: hybrid retrieval (BM25 + vectors), citation graphs, and domain ontologies to improve recall and precision.
• LLM-assisted workflows: screening, classification, PICO extraction, summarization, and bias/quality flags with human-in-the-loop review.
• Evidence provenance: source links, quote-level citations, dataset versions, and audit logs baked into outputs and APIs.
• Evaluation you can trust: gold sets for screening and summarization, judgment consistency checks, and continuous quality monitoring.
• MLOps for "living" updates: scheduled crawls, event-driven pipelines, drift detection, rollback strategies, and cost controls.
• Security and compliance: handling publisher licenses, PII redaction, role-based access, and jurisdiction-specific retention rules.
• Delivery: stable APIs, SDKs, and lightweight UI components so policy teams can query, compare, and export evidence quickly.

How the system might work

Think pipeline: ingest research → enrich with entities, topics, and quality signals → index in a hybrid store (keyword + vectors) → orchestrate LLM steps for screening and synthesis → route to reviewers → publish with citations and versioning.

Practical stack choices could include a vector database, an orchestration framework for tool/LLM calls, a workflow engine for review loops, and a feature store for reuse. Knowledge graphs and citation networks can lift recall and explainability without slowing response times.

Pilot domains and use cases

• Education: "What interventions improve literacy within 12 months for ages 7-9?" with cost and context filters.
• Justice: evidence summaries on diversion programs, with confidence scores and implementation notes.
• Environment: climate adaptation measures ranked by effect size, locality, and time-to-impact.

Partners, funding, and scope

METIUS is led by Queen's University Belfast with the University of Exeter, working alongside the University of Newcastle and PenARC. It is funded by UK Research and Innovation through ESRC and NERC, with co-funding from the Department for Science, Innovation and Technology (DSIT). Full details are available via UKRI: AI investment to transform global policy with scientific evidence.

What to do next

If your team is building evidence pipelines, now is a good time to sharpen skills in retrieval, LLM orchestration, and evaluation. You can explore focused upskilling here: AI courses by job.