Cambridge's latest AI-deas Sprint cohort will spend six months building and testing tools with domestic-abuse practitioners, industrial partners, local decision-makers, manuscript scholars and botanic-garden curators. The five projects, announced on 22 October 2026, share a practical task: turning complex information into something specialists can use.[1]
The programme provides funding and technical support from ai@cam's machine-learning engineers. Several projects build on existing tools or prototypes, with the sprint intended to establish how they work for prospective users. Cambridge's announcement sets out the development plans and partners; operational outcomes will follow from those tests.[1]
Five projects with different definitions of success
A project led by Hannah Pankow will work with Cambridge Women's Aid and lived-experience advisers on technology-enabled domestic abuse. Its practitioner-facing tool is intended to map possible digital exposure and support a staged safety plan, bringing technical, legal and safeguarding knowledge into the same workflow. It is designed to support practitioners' judgement and identify when specialist help is needed.[1]
Second Ahead, led by Tyler Holderness and colleagues, will initially focus on forklift operations. Standard video feeds will be used to explore whether gaze, pupil diameter, blink rate and heart-rate variability help identify emerging risks around machinery. Raw video is to be processed locally and discarded, while the project investigates workplace-privacy questions alongside predictive performance.[1]
The five Cambridge prototypes
| Project | Work planned | Users and setting |
|---|---|---|
| Digital-abuse support | Map exposure and help stage safety plans | Cambridge Women’s Aid and lived-experience advisersSource 1 |
| Second Ahead | Explore early warning signals around machinery | Initial focus on forklift operationsSource 1 |
| Air-quality mapping | Estimate street-level nitrogen dioxide and uncertainty | Greater Cambridge PartnershipSource 1 |
| Manuscript reconstruction | Suggest fragment matches for scholarly confirmation | Initial focus on the Cairo GenizahSource 1 |
| BGsmartR Curator | Add a conversational interface to collections analysis | Botanic gardens across four continentsSource 1 |
Dr William Lee's air-quality project will combine atmospheric-dispersion modelling with sensor measurements to estimate nitrogen-dioxide concentrations and their uncertainty at street level. Working with the Greater Cambridge Partnership, it will examine the possible effects of extending 20mph limits on selected roads. Weather and traffic variation make this an exercise in separating the intervention's effects from other changes.[1]
Ancient manuscripts pose a different validation problem
Vince Beiler, John Dudley and colleagues will expand an existing prototype for matching manuscript fragments. The initial collection is the Cairo Genizah, with around 350,000 medieval Hebrew and Judeo-Arabic fragments, roughly 60% held by Cambridge University Library. Scholars will inspect suggested matches and record their decisions in a shared workspace, providing feedback for subsequent recommendations.[1]
BGsmartR Curator, led by Professor Sam Brockington, will add a conversational interface to an existing open-source analysis tool. Its target users are curators who need to ask about conservation priorities or vulnerabilities in living collections without specialist data expertise. The project will involve gardens across four continents and Botanic Gardens Conservation International.[1]
Specialists shape what counts as a useful result
The projects make feedback part of development: scholars confirm a fragment match, practitioners judge a safety-planning aid and curators test an answer against collection records. That gives each team a specific way to investigate usefulness beyond a convincing demonstration.[1]
The cohort puts specialist judgement inside the development process. A scholar's confirmed match, a practitioner's assessment of a safety plan or a curator's check against collection records can each provide concrete feedback while the prototype is still being built.[1]
Sources & further reading
- From safer workplaces to medieval manuscripts: five new AI-deas projects put AI to work on real-world problems University of Cambridge · 22 Oct 2026 · Accessed 27 Oct 2026