From transforming brain tumour surgery to helping millions of smallholder farmers protect their crops, seven new AI-deas Sprint projects will use artificial intelligence to tackle some of society’s most pressing challenges. The teams, successful in a competitive open application process, will receive funding and technical mentoring to develop AI prototypes addressing real-world problems across healthcare, education, climate, and urban planning.

The projects span an ambitious range of challenges: making conservation evidence instantly accessible to environmental decision-makers, developing diagnostics for women’s health, revolutionising brain tumour surgery with thermal imaging, automating crop disease warnings for food-insecure regions, creating personalised second language learning tools, improving pelvic health through AI-powered biofeedback, and helping UK communities understand local housing developments. Each represents a practical test of how cutting-edge AI research can deliver tangible societal benefits.

“The AI-deas Sprint Programme is about bridging the gap between promising research and real-world impact,” said Jess Montgomery, Director of ai@cam. “These seven projects demonstrate the breadth of challenges that AI can help address, from improving women’s health outcomes to supporting global food security. What makes this programme special is that we’re supporting interdisciplinary teams to move beyond traditional research and build working prototypes that can attract partnerships and funding.”

The seven funded projects address diverse challenges across multiple domains. The projects are:

• Conservation Co-pilot: Making Environmental Evidence Accessible

With over 1 million users already accessing the Conservation Evidence database, this project will develop the first ‘Conservation Co-pilot’ – an AI-powered chat interface that retrieves, summarises, and presents conservation evidence to answer user questions. The challenge is ensuring that AI faithfully represents scientific evidence without misrepresentation. Building on rigorous evaluation research comparing frontier AI models with human experts, the team will create an agentic system that draws together evidence while maintaining faithfulness to source material. The tool will be game-changing for conservation decision-makers seeking evidence to guide action.

Team: Sam Reynolds (Department of Zoology) – Project Lead; Alec Christie (Department of Zoology); Sadiq Jaffer (Department of Computer Science and Technology); Anil Madhavapeddy (Department of Computer Science and Technology); Rebecca Smith (Department of Zoology); William Sutherland (Department of Zoology)

• Dr. Endo: Accelerating Endometriosis Diagnosis Through AI-Enhanced Ultrasound

Endometriosis affects 1 in 10 women, causing severe pain and fertility complications, yet diagnosis typically takes 7-10 years. Between 35-50% of people struggling to conceive naturally suffer from endometriosis, but detecting the condition requires specialist expertise and it takes sonographers 6-9 months to gain proficiency in identifying the five key sonologic signs. This project will develop an AI tool that analyses 2D ultrasound scans to detect subtle patterns of endometriosis, enabling earlier and more consistent diagnosis. The system will generate grounded outputs linking text descriptions to visual regions of interest, creating bounding boxes for features like endometriomas and adhesions. Beyond improving diagnosis, early detection can prevent cyst development, ease debilitating symptoms, avoid expensive invasive procedures like laparoscopies, and significantly improve IVF outcomes.

Team: Staci Weiss, Mo Vali and May Levin

• Thermal Imaging for Precision Brain Tumour Surgery

Brain tumours are the leading cause of cancer death in children and adults under 45, with complete surgical resection being the most powerful predictor of survival. Distinguishing between healthy and tumour tissue during operations is challenging, because of how the brain naturally shifts during surgery. This project will advance a novel passive infrared imaging system that can overcome this challenge. Building on a successful first-in-human feasibility study at Addenbrooke’s, the team will systematically define the technology’s precision at tumour boundaries through controlled laboratory and preclinical studies. The research will establish the evidential foundation for NHS clinical trials of a cost-effective, label-free surgical tool with potential to transform brain cancer outcomes.

Team: Stephen Price, Clinical Professor of Neurosurgical Oncology at Department of Clinical Neurosciences; Richard Mair, Assistant Professor and Honorary Consultant Neurosurgeon at CRUK Cambridge Centre; Gita Khalili Moghaddam, Principal Investigator at the Department of Clinical Neurosciences.

• AI-Powered Crop Disease Advisories for Global Food Security

With climate change accelerating crop diseases and threatening food security for millions of smallholder farmers across Africa and South Asia, this project will deploy AI to automate the generation of wheat disease risk advisories. Currently, expert pathologists manually write and translate weekly summaries of disease forecasts. This labour-intensive process limits scalability as demand grows. The team will fine-tune a large language model on five years of expert-edited advisories, enabling it to generate concise, context-aware summaries tailored to each country’s communication channels. The system will be augmented with real-time disease forecasts and survey data, dramatically increasing the responsiveness of early warning systems that currently support international collaboration spanning nine countries and tens of millions of farmers.

Team: Jacob Smith (jws52@cam.ac.uk), Research Associate, Department of Plant Sciences; Lawrence Bower (lb584@cam.ac.uk), Computing Research Associate, Department of Plant Sciences; Chris Gilligan (cag1@cam.ac.uk), Research Professor, Department of Plant Sciences.

• Personalised Second Language Learning Through AI

Despite the societal and economic benefits of second language learning, participation in language education has been declining, with marked inequalities in provision across educational settings. This project will deliver an AI-based training tool that provides personalised feedback to enhance language education. Building on research with over 200 adolescent learners that mapped cognitive and environmental predictors of language proficiency, the system will assess each learner’s phonological working memory, awareness, and aptitude to generate personalised training pathways. A compact language model will provide tailored feedback ranging from articulatory guidance to perceptual cues, adjusting practice intensity based on learner strengths. Beyond enhancing outcomes, the tool promises to make language learning more accessible and inclusive across diverse educational contexts.

Team: Mirjana Bozic (Department of Psychology), Brechtje Post (Department of Theoretical and Applied Linguistics), Linda Bakkouche (Department of Theoretical and Applied Linguistics)

• PelviSense: Non-Invasive Biofeedback for Pelvic Floor Health

Pelvic floor muscle dysfunction affects over 14 million people in the UK, costing the NHS more than £230 million annually. While pelvic floor training is the recommended first-line treatment, its effectiveness is limited by restricted physiotherapy access and the lack of feedback to guide correct technique. This project will transition a laboratory prototype for non-invasive pelvic floor monitoring into a population-verified tool, eliminating the need for uncomfortable internal probes. The team will expand data collection, develop next-generation algorithms that improve accuracy and ease-of-use, and conduct extensive user validation to ensure the system is reflective of community rehabilitation settings and aligned with NHS needs.

Team: Dr Hristo Dimitrov (MRC Cognition and Brain Sciences Unit), Ms Alexandra Williams (MRC Cognition and Brain Sciences Unit), Prof Tamar Makin (MRC Cognition and Brain Sciences Unit)

• PLATO: Making Housing Development Transparent

The UK housing crisis is characterised not only by shortage but by information asymmetry – residents lack clear, integrated information about what is being built, where, and whether schools, healthcare, and green spaces will be sufficient. This project will deliver an AI-driven platform that retrieves planning documents from Cambridgeshire councils, classifies content using a fine-tuned 120-billion-parameter model into ten policy domains, and generates evidence-backed summaries with exact source citations. The interactive map will visualise proposed developments, identify amenity gaps, and assess climate exposure, enabling both planners to confirm schemes meet guidance and residents to understand what will be built in their communities. By reducing information asymmetry and providing transparent access, PLATO aims to improve planning efficiency while strengthening public trust and engagement.

Team: Dr Jerry Chen, Department of Land Economy; Emily Tianyuan Wang, Department of Land Economy; Prof Li Wan, Department of Land Economy

From Ideas to Impact

Over the next six months, project teams will develop and test their prototypes with support from ai@cam’s technical clinics and mentoring. They will share their findings at a Demo Day in May 2026, highlighting lessons learned and pathways for wider adoption, partnerships, and follow-on funding.

” By supporting interdisciplinary teams to build working prototypes grounded in real challenges, we’re demonstrating how AI research can deliver genuine public value,” said Professor Neil Lawrence, Chair of ai@cam. “Universities have the interdisciplinary expertise, the research infrastructure, and crucially, the freedom to tackle challenges that matter for public good. These seven projects demonstrate how universities can be engines of responsible innovation – developing AI solutions grounded in both cutting-edge science and real-world need.”

The AI-deas Sprint Programme is part of ai@cam’s mission to drive AI innovation that makes a positive difference to society and delivers real public value.