AI Risk Decoder
Client - Edyta Bogucka, Nokia Bell Labs
AI systems are increasingly deployed in sensitive domains such as hiring, healthcare, and policing — but the risks they pose are often hidden deep inside compliance documents, risk registers, or opaque terms of service. Legal Design seeks to make these risks visible, understandable, and actionable for non-experts. Your task is to create an AI-powered assistant that can extract, interpret, and redesign how AI risk information is communicated. The system should take legal or policy text (e.g. risk assessments, safety standards, liability clauses) and transform it into accessible outputs: personalised summaries, visual diagrams of risk flows, or “at-a-glance” dashboards for different stakeholders (citizens, regulators, companies). The challenge lies in combining natural language processing, risk modelling, and user-centred design to prototype a tool that makes AI accountability tangible.
Binge Smarter
Client - Michele Benciolini, Bending Spoons
Many viewers enjoy international films, series, and online content that they watch with English subtitles, or English-language content with subtitles to help learn vocabulary and idioms. At present, recommender algorithms in Netflix or YouTube give no assistance with following the plot, keeping track of which characters have appeared, or helping assess whether you are learning new language skills. This project will create a service that uses natural language processing APIs to match content to language learning, for a more linguistic viewing experience.
Clinical Trial Agent
Client - Stephen Baker, Evinova
Design and prototype a web-based system that optimises clinical-trial visit schedules for patients and trial sites using only synthetic data and public APIs. The system ingests a trial protocol (visit windows and constraints), patient availability, and site capacity, and produces feasible schedules plus simple risk flags (e.g. missed visit windows or conflicting medications). The prototype demonstrates calendar export (iCalendar / .ics) and an optional integration with a single demo calendar service. An optional stretch goal is to showcase an “AI computer-using agent” operating in a sandboxed environment to automate a scheduling task.
Coding for Blind Learners
Client - Richard Pawson, Naked Objects Group
A modern IDE such as VS Code is ‘accessible’ to blind users (with a screen reader and keyboard commands) – and some blind professional developers are very proficient with them. But the design of such IDEs still reflects how sighted users think and work, adding to the challenge for blind programmers, while failing to leverage their unique strengths. Your task is to create an open-source tool specifically to help a blind person learn to program from scratch, in a text-based language. The tool is not expected to be attractive to sighted users, though it must still be usable by a sighted teacher/helper. The client will arrange for one or more blind users to provide feedback during the project.
Cold Bridge Diagnostics
Client - Bart Hommels, Open Eco-Homes
In a well-insulated house, any continuous path of brick or stone can become a “cold bridge” that leaks indoor heat on cold winter days. The goal of this project is to allow a home-owner to diagnose this problem through time-lapse analysis of photographs taken with thermal imaging cameras at different times of day and seasons. You’ll need to use computer vision methods to normalise and register infrared/optical photographs that have been taken from slightly different conditions and poses, and then construct a heat flow model from differential analysis of the image regions.
Community Culture during Diaspora
Client - Benaiah Matheson, Carriacou Historical Society
The cultural life and livelihood of Carriacou people was once shaped by a didactic practice — building and sailing wooden schooners, farming, and collective self-reliance on the island. These traditions marked everyday life in the mid-20th century, before many migrated to the UK and US, and continue to inform how Carriacou is remembered as home and what it has become today. Your design challenge is to create a compelling experience of this heritage, both online and for exhibition as mixed/augmented reality. Layer 1 represents the Carriacou community of that era, who created and held this space. Layer 2 invites contributions from the diaspora with permission. Layer 3 is observational, offering outsiders abstracted representations. Oral histories include dialect words and accents unrecognised by mainstream AI, preserving their links to tools, materials, ecosystems, and crafts. The visual rendering and interaction style should draw from Carriacou traditions such as boatbuilding, sea voyages, Maroon, or the calabash tree, treating them as living systems of knowledge rather than decorative motifs.
Detecting Patterns of Rhetoric
Client - Christopher Newfield, Independent Social Research Foundation
Some styles of writing can be recognised by patterns of similarity, even when different words are used. Think of physical actions, where any one kick or wave might be different, but the general characteristics of kicking or waving can be learned, given enough training data. Microsoft's Kinect was trained in part by CGI-generating a huge number of synthetic examples of kicks, waves and other gestures. Your task is to make a detector, not for physical gestures, but for certain ways of writing, by training a detector using huge amounts of synthetic writing generated by an LLM. Wouldn’t it be amazing if we had an automatic bullshit detector, which could be trained to detect human BS by showing it enough examples of artificial BS from ChatGPT?
Digging for Data
Client - Leontien Talboom, Cambridge University Library
In the century before the World-Wide Web, a whole generation of scientists kept their data and archives on floppy disks, often encoded in formats specific to a particular disk drive or PC manufacturer. Much of that data has never gone online, and is now in danger of being lost forever. This project is an archaeological exercise, developing a tool that can excavate and decode that data, perhaps applying machine learning methods to identify and reverse-engineer texts and tables from the magnetic media.Your client is a digital archive specialist, who can provide many samples, as well as technical documentation, to get you started.
Draught Detective
Client - Fran Sutton, Cambridge Carbon Footprint
A key priority in making homes more energy efficient is to find and plug air draughts. In principle, a builder upgrading an existing house could simply fasten a fan in an open window, use a differential pressure meter to determine how airtight the house is, and start plugging holes, adjusting the number of air changes per hour to fall within a desirable range. A cheap and easy phone app could help the builder to setup and calibrate the equipment, then provide practical guidance to prioritise the most urgent repairs, giving feedback on what has been gained with each of those.
Drawing Machines
Client - Pablo Garcia, Art Institute of Chicago
The ancestors of computer graphics, in the 18th and 19th centuries, used sophisticated mechanisms to render realistic perspective views, transform complex geometry, visualize mathematical and real-world data, and act as human prosthetics. Pablo Garcia’s drawingmachines.org documents many of these remarkable devices, but we have lost much of the information on how they really work. The goal of this project is to build an interactive assembly kit that simulates mechanical drawing machines, bringing them back to life, and even providing a creative resource for today’s art students.
From AI to BI
Client - Philip Turon, Cambridgeshire Beekeepers Assocation
Beekeeping is valuable for the environment, an enjoyable activity in many Cambridge colleges, and makes delicious honey. But honeybees in the Cambridge area are at threat from diseases that could be better detected and controlled, if only there were more bee inspectors (BI's)! Your task is to create a mobile application that supports this community, including use of AI to recognise some serious diseases from photos, collects data for bee research, and guides beekeepers to make better use of the tools and support available.
Futures Radar for Education
Client - Kevin Martin, Digital Education Futures Initiative
The UNESCO Futures of Education observatory asks how the world can renew education, bringing new ideas as well as public debate to inspire research and action. Your task will be to create a futures thinking tool that collates and visualises the signals of change, in an open-access interactive tool that can be made available to policymakers, teachers, school leaders, and students around the world. Combining public source data, structured datasets, and project-specific inputs, the Futures Radar will use open-source visualisation techniques to explore maps and timelines, alongside visualisations of uncertainty, as a platform to support more informed and resilient decisions.
Intuitive Business Data
Client - Ben Schofield, Calero
Visual programming languages like the educational product Scratch could be more accessible for nontechnical business users, but most genAI coding assistants output Python or Javascript that is hard for non-programmers to read and adjust. Your task is to create a novel programming tool for business data processing, where the first draft of the solution is generated from a natural language text prompt, but users can refine their solution just by selecting and dragging graphical data processing elements rather than endless chatbot interaction.
Interactive Assembly Visualisation
Client - Hugo Pompougnac, INRIA Grenoble
Performance engineers often work with assembly code, whether it's generated by a compiler or written by hand. Yet, such code is hard to read -- and therefore hard to write. Writing such code becomes easier if the presentation of assembly code is visually enhanced: masking overly low-level details (like instruction addresses), annotating control flow, syntax highlighting, and even lifting the representation of the computation. In this project, you will integrate such techniques into xDSL, a general-purpose MLIR-compatible compiler framework written in Python with representations of assembly instructions, to help with compiler debugging as well as debugging handwritten assembly. This will involve: parsing the assembly, converting it into MLIR-structured IR, as well as creating a visualisation of useful information present in the assembly. We invite you to take inspiration from anywidget, textual, or any other related tool you may find. When completed, this project will make interactive assembly code reasoning easily accessible and fun.
Junior Space Engineers
Client - Thomas Garnier, French Space Academy
Many space programmes offer popular opportunities for schools to propose and deploy experiments in space. This project provides a way for kids to design proposals for their own space hardware. IDM CIC is a systems engineering plus CAD tool developed by CNES, the French Space Agency. It wasn’t created for use by children, so your task is to develop an accessible design assistant, that can start from a voice prompt to explore “vibe” design ideas, then lead them in to the physics and engineering detail of defining a deployable satellite module.
Local Medical Guidelines
Client - Charlie Artingstoll, Medical Action Myanmar
In many parts of the world, medical guidelines are created specifically for local contexts, but get published only as PDFs, not accessible when and where they are needed - on low-powered devices with poor network connectivity. The goal of this project is to make an authoring system that easily allows local medical teams to create, distribute and maintain medical guidelines as a lightweight progressive web app, ideally drawing on AI enhancements and community knowledge to combine the best features of Wikipedia and ChatGPT.
Nanosis
Client - Ljiljana Fruk, BioNano Engineering Group
Inspired by Tetris-style games, you will design Nanosis. Nanosis will be a game where players assemble nanoparticles, molecular linkers, and biomolecular ligands (peptides or DNA) to overcome biological barriers and deliver therapeutic payloads to target cells. Each level represents a stage in a nanomedicine’s journey through the body: from circulation in the bloodstream, where players must avoid immune cell uptake, to crossing blood vessel walls, navigating the tumour microenvironment, and ultimately achieving cell entry and intracellular release. Nanoparticle design integrates realistic, structure-informed challenges inspired by protein models (e.g. AlphaFold data on cell receptors) and existing building blocks ( lipids for lipid nanoparticles, polymers for polymeric), helping player appreciate how nanoscale design choices (e.g., particle size, surface chemistry) influence biological interactions. The game aims to spark curiosity about nanotechnology, molecular engineering, and future medicine, while reinforcing concepts such as biocompatibility, targeting specificity, and controlled release.
R U NOT HUMAN
Client - Sarah Ciston, Academy of Media Art Cologne
Are CAPTCHAs just free labour for AI-training? The internet is in an arms race between CAPTCHAs and bots that solve CAPTCHAs, but only humans are losing out. Investigative artist Sarah Ciston has won prizes for exposing military use of AI. This team will work with her to design a system that filters unwanted web traffic without judging people as "not human enough", for example if they use assistive technologies like screen readers or gaze trackers. A public demonstration will show what really happens when humans get captcha'd by social media and surveillance tech.
Soundscapes of Belonging
Client - Edyta Bogucka, Nokia Bell Labs
Mobile phones can augment our everyday experience of place, but too often they pull attention away from the environment rather than deepening our connection to it. This project asks: how can context-aware mobile technologies foster place attachment without being intrusive? Your task is to design and prototype a mobile app that plays short, tailored audio sentences (around 5 seconds each) that resonate with the user’s immediate surroundings. Drawing from large-scale datasets like HappyDB and spatial data from OpenStreetMap, the system will select and adapt sentences to reflect specific types of urban spaces (such as parks, plazas, or streets). You will then evaluate whether these micro-audio experiences can meaningfully enhance people’s sense of belonging and emotional connection to their physical environment.
System Verilog for CIRCT
Client - Moritz Scherer, Mosaic
Open-source hardware development is rapidly gaining traction, with projects such as RISC-V, CHIPS Alliance, and CIRCT aiming to make hardware design as transparent and collaborative as open-source software. A key challenge is building reliable, open compiler infrastructure that can handle the complexity of modern hardware description languages like SystemVerilog. This project contributes to that effort by extending the CIRCT-Verilog compiler so that a wider range of real-world SystemVerilog examples can be compiled into CIRCT’s core MLIR dialects. The team will identify unsupported constructs, implement new compiler transformations, and validate their work by expanding the set of examples that successfully lower through the CIRCT toolchain.
The Admissions Game
Client - Rachel Thorley, STEMfluent Ltd
Cambridge admissions interviews, like the college supervisions they could lead to, are all about conversation. Yet high school students with no personal tutor or private school may never have had a human conversation about the subject they love. Some might resort to a chatbot, but this project is about making human connections instead, perhaps exploring topics from an online resource like Isaac Physics. Pairs of users can practice interviewing each other, with an algorithm like TrueSkill maintaining scores to help students from all kinds of background assess their chances.
Video Delphi
Client - Ric da Silva, Flok.health
Social media and video streaming platforms create algorithmic bubbles that polarise public debate in unhealthy ways. In contrast, the Delphi Technique is used in healthcare and other domains for reaching consensus, by iteratively presenting a panel of experts with a question. This project is to build a media platform for developing LLM-assisted Delphi consensus. The platform should enable video and text forums to be set up and configured, perhaps in the style of Substack. But also, importantly, it should include comprehensive introspection and analytics on the discussion process as it unfolds so that the configuration may be updated to improve the result in a subsequent run.
Wise Banking
Client - Tokunboh Ishmael, Alitheia Capital
Mobile banking apps are convenient, and help many students with monthly budgets, but can make it hard to plan for the long term. A particular failing is how to get life advice from other generations - not just parents, but wise friends and mentors in your home community. Your challenge is to create a personal wisdom-based banking app that connects across generations, drawing on principles from the Yellow Cowries curriculum that empower young adults to make improved financial decisions such as savings, loans, investments and insurance. It’s important to consider and go beyond AI advice models, building whatever is special in local communities and lifestyles.
