skip to content

Department of Computer Science and Technology


Acoustic Land Management

Client: Adham Ashton-Butt, BTO

Land managers are often responsible for restoration projects, in which they make changes expected to benefit wildlife over time. One very useful source of evidence is acoustic data recordings. This project is intended to create an interactive application that can be used by land managers to visualise, make decisions, monitor, and evaluate their interventions, based on bioacoustic survey data processed through the BTO acoustic pipeline and enhanced with machine learning methods.

Braille Predictive Text

Client: Gregory Hargraves, Paige Braille

David MacKay’s Dasher was a radical predictive text system where a machine learning language model helped users with disabilities by guiding them to zoom in to more likely possible texts. That’s no use for people with a vision impairment, who can’t see the guidance or easily operate animated graphical interfaces. This is an opportunity to work with a team who have recently augmented the classic mechanical Braille typewriter with internet connectivity. Your prediction model will help users to learn and master Braille, by offering appropriate spoken suggestions on what word they might be typing and (if needed) remind them which keys they might need to press next. All this will have to be designed in a way that is helpful, motivational and accessible, not an additional obstacle to learning Braille!

Climate Foresight

Client: William Jones, Embecosm

Dynamic Causal Modeling is a Bayesian statistical technique for reverse engineering time series data. One of the ongoing challenges in applying such statistical models is how to visualise the multiverse of possible outcomes that the algorithm derives. Your goal is to create an evidence-based visualisation of possible climate futures that allows users to interrogate and compare projections from a complete simplified carbon-climate model within the Dynamic Causal Modeling framework.

Component Quest

Client: Stephen Devlin, Sensors CDT

The electronics industry faces chip shortages, rising costs, and increasing e-waste. This creates demand for secondhand components, but it can be difficult to identify and value them. Develop a smartphone app that uses AI to identify and value electronic components (usually still mounted on PCBs) using the phone camera. The app should handle a wide range of components, provide current market value, and link users to a marketplace to buy and sell them.

Copilot for Business

Client: Jason Mashinchi, Cambridge Kinetics

We’re all familiar with the code generation capabilities of ChatGPT and GitHub Copilot. That’s fine for experienced programmers, but how could an AI coding tool support people who have never written a line of code? Your goal is to create a user friendly natural language interface that not only generates code for extracting and visualising data from an arbitrary business database, but helps non-programmer users to understand how that code works, recognising and correcting bugs or hallucinations in the generated code.

Creative Writing Coach

Client: Rachel Thorley, Inkli

Will human authors be replaced by AI? Language models can convincingly generate human-sounding text and automate many writing tasks, even contributing to novels! Instead, could they be harnessed to make us all more creative? The goal of this project is to create an app in the style of Duolingo. This app will use bite-sized, gamified lessons to help students write more creatively, with poetic imagery, dramatic tension, and vivid characterisation. Experiments show that a standard LLM (with carefully designed prompts) can provide useful feedback. Your app needs to engage and motivate students to become more fluent and confident creative writers.

Disability Bias Explorer

Client: Cecily Morrison, Microsoft Research Cambridge

Recent work led by a team at ethical AI initiative Hugging Face (arXiv:2303.11408) demonstrated a tool that could be used to interactively investigate some of the race and gender biases that have become encoded in major text to image generation systems. Your task is to create a similar tool that could be used to demonstrate image generation biases relevant to visual disabilities. You may also need to consider how this tool can be made accessible to users who have visual disabilities themselves.

DoS D-Stress

Client: Simon Fothergill, Cam AI

The very first AI chatbot was the psychotherapist Eliza. For young people today, mental health is more challenging than ever, but we aren’t using LLM technology in ways that reduce stress. Your goal is to create a de-stressing experience, potentially including responsive sound and imagery alongside short generated text exchanges that respond to what is stressing you. A fully automated chain should direct stressful content (e.g. an email from your DoS) directly into the system so that you can be properly de-stressed before you even read it.

Engaging Everyone

Client: Rachel Gardner, Communications Manager

Many public events and services aim to include young people, with meaningful attention to their specific views and needs. Our own outreach activities, such as the huge Cambridge Festival (next taking place on 16th March 2024) want to ensure that feedback from visiting families includes all visitors, especially where “family” feedback might otherwise only engage adults through traditional surveys. Your task is to create a fun and engaging online tool that encourages people of all ages to share their views, and is inclusive and motivational for the young. If successful, we can deploy this at the Cambridge Festival as soon as it is finished.

Envisioning Nairobi

Client: Nkatha Gichuyia, Urban Re-engineering Taskforce

Large cities around the world are grappling with plans to accommodate environmental, social and climate change. But there is a huge gulf between the technical software of geographical information systems and infrastructure planning on one hand, and the retail and entertainment data that is promoted in consumer products like Google Maps. The city of Nairobi is thinking ahead through visionary planned cities as Nairobi Railway City, Tatu City, and the Konza Technopolis, that aim to present models of how an environmentally sustainable and climate-resilient urban area developments in a carbon-constrained environment can be delivered. Your role is to prototype a new social mapping tool that could engage members of the local population in these discussions, visualising data, plans, priorities, and their personal implications.

Heterodox Economic Modeller

Client: Christopher Newfield, Independent Social Research Foundation

There is a long history of financial simulators that can be used to model national economies, dating back to the water-powered “Phillips machine” (we have one in Cambridge). Today’s economy is stuck in a rut, because alternative economic theories are proposed by people whose mathematical models are simply less well embedded than the legacy implementations. Your task is to create a quick and dirty financial simulator that could be used by heterodox economists to propose and debate serious alternatives to outdated government and corporate policies.

Influencing Health

Client: Lee Wilson, Kavavar

Many more people get their news from social media than TV, but much national and international public health policy relies on public broadcasting as the primary channel of communication. Your job is to make a tool that helps reconnect local health emergencies and interventions to local influencers who may be motivated to help. This will involve using APIs from TikTok, Instagram etc, but mapping these against local health statistics in a way that doctors and government officials can see what is going on.

Investment Provenance

Client: David Friedman

There are many online forums that offer investment advice, but not much of it is original. Your task is to create a tool for use by traders, that annotates investment tips with a plausible history of where they might have come from. Any correlation with historically objective information, such as share price movements, may be a helpful source of causal evidence. This project is likely to require some research into text processing algorithms and statistics, as well as an intuitive front end that helps users understand the "genetics" of specific investment trends and memes.

Memories Retold

Client: Jonathan Sunderland, Ab Initio

Design and develop an application to help people with memory loss create an oral history through interactive prompted storytelling. The application should use AI to identify suggested talking points and ask contextual questions to trigger the unlocking of long-term memories. The application should accept images as input prompts, and use speech-to-text conversion to capture the user's responses. Ethical aspects should be considered, including who might use the application, in which circumstances, and when.

Optimising Music Notation

Client: Arild Stenberg, Score Designs Group

The LaTeX typesetting system can be used to render many alternative styles of document from a master text specification. In contrast, music typesetting systems like Sibelius make all music look much the same. Your client has designed an optimised alternative style for music notation that has been found to improve performance. Your task is to define a specification language for these modifications, and implement a rendering pipeline either as Sibelius extensions, or using an open source music rendering back-end such as PMW (whose author, Philip Hazel, will be available as a consultant to the project).

P2P Social Network

Client: Mick Vermeulen, IMC

Social networks are a major part of our lives. The acquisition of Twitter by Musk demonstrates how much influence one person can express by changing the platform to his liking where users have no say in the direction of the platform. Your assignment is to create a peer 2 peer honest social network: a social network that has no central server that anyone can join freely. You'll need to find or define a protocol that allows users to communicate. Will you go truly peer2peer? Or would you prefer a node-based approach where any server can join the network and serve requests for users? It's important to ensure that users cannot imitate each other. How do you prevent a hostile actor from joining the networking and pretending to be someone else? There's no central authentication server to verify this. Your solution should include a GUI that allows for browsing user accounts. How would this work? How do you fetch information about a user if there's no central server to provide that, and should there be any restrictions?

Ramping Up Sustainable Crops under Climate Changes

Client: Ji Zhou, NIAB

Arable crops in the countryside around Cambridge rely on soil nitrogen (N), but many crop varieties are inefficient in the uptake of this critical resource, leading to the increase of farm costs and other potential environmental risks. Your goal is to use advanced vision-based artificial intelligence (AI) techniques to analyse satellite and drone-collected imagery to implement algorithms to assess the use of N fertilisers from the field experiments. Real world datasets collated by the Data Sciences Department at NIAB will be provided to help establish training datasets, but there may also be possible to engage with ongoing trials in NIAB-Cambridge to verify your findings, so that patterns of genotype-environment interactions (GxE) under different N treatments can be revealed. [note: following a change of client at NIAB, this design brief has been modified slightly since the original advertisement]


Client: Sandi Toksvig, Mappa Mundi project

The training data for AI large language models mixes up Wikipedia articles with random text from social media trolls and commercial sponsors. Many voices are excluded, from the girls not being educated in Afghanistan, to the women who teach their own families all over the world. Your goal is to make an alternative wiki architecture, where the benchmark of authority is not who has the most money or followers, but the best reflection of those voices. You’ll need to think about what motivates women to spend their time this way, what kind of devices they have access to, and what makes knowledge reliable. Your prototype might be a small start, but the AI of the future depends on a better way of understanding what is meaningful, and Wikipedia founder Jimmy Wales should be available to discuss this with you.

Supply Chain Resilience

Client: Murad Abdulla, IMC

Events such as the Covid pandemic and the Ukraine-Russia war highlight inefficiencies in global supply chains, affecting our everyday lives through energy prices, inflation, cost-of-living, unemployment and housing. This project will use publicly available datasets and APIs to build a comprehensive map investigating supply chain choices such as semiconductors for tech via China to the West, oil exports, agricultural product shipments etc. You will create metrics for supply chain efficiency and resilience, ranging from route stability to sensitivity to natural disasters. This visualisation should offer opportunities to optimise routes for delivery of different industrial products, contrasting and comparing CO2 emissions, cost, reliability, and robustness through diversity of supply options. It will be important to ensure that data is not biased, and that features are explainable, to support public accountability for decision makers.

Talking Music

Client: Cecily Morrison, Microsoft Research

People with different sensory capabilities or diverse neural styles sometimes find it easier to communicate via a musical instrument or a piano keyboard than with the screen and keyboard of a laptop. The goal of this project is to make an alternative kind of music editor that works with text and sound as the elements, rather than working with music notation on the computer screen. The open source synthesiser SuperCollider generates professional quality sound, but you would replace the sclang programming language with a natural language interactive text system that could be driven from an assistive keyboard or screen reader.

Taste: Movies x Books x Music

Client: Hrvoje Abramović, IMC

There are many music, movie and book recommendation systems online, but each focuses on a single medium. The goal of this project is to build a web application which, when given a movie or a book, would present a cluster of books, movies and songs that are related to it by often being mentioned online together. It would be cool to see how clusters of “tastes” form and how books, movies and music from the same artistic movements naturally pop up together. A timeline could also give an idea of the art history context, for example the movie Easy Rider might relate to movies, books and songs popular with the hippy movement.

Testing for Humans

Client: Bhasi Nair, Equitech Futures

Many companies rely on online programming aptitude tests to select potential employees. Unfortunately, applicants can often fool the test by finding answers online, or even generating a response automatically using an AI chatbot. Your task is to create an authoring and administration tool for aptitude question banks that automatically benchmarks them against such strategies, helping the question author to fine-tune them in a way that will guarantee human responses.

Visiting the Forest Stream

Client: Ana Cuza, ClimateInColour

There are many Spotify playlists featuring the sound of nature, but it's hard to visit the places those sounds were recorded, or meet the people that live there. This project is an environmental update on music artists who make more money from concert visits and merchandise than they do from streaming licenses. Your client has been working with a forest community in Ghana, where the environmental soundscape doesn't bring any revenue to the village. The goal is to create a geolocated mobile app and business model, connecting distinctive sounds to new revenues from ecotourism.

Wearable Sleep Coach

Client: Alex Venetidis, Terra

Terra API unlocks health and fitness data from diverse sources, including popular wearables and fitness apps like Fitbit, Apple Watch, Oura, and 70+ others. Your challenge is to leverage the Terra API to develop an AI coach that advises people on how to improve their sleep habits. The coach can be trained using data already acquired by Terra. Throughout the process you will be prompted to deploy and evaluate the performance with real users, do user interviews, and iterate on your product with the goal of attaining 100 users - this stage will also involve ethical review for deployment of instrumented software.

World Craft

Client: Sandi Toksvig, Mappa Mundi project

Google Earth, Google Maps and Google Streetview have turned surveillance data into commercial opportunities ranging from Deliveroo to Down My Street. But coverage tends to focus on sponsored business opportunities, rather than the livelihoods and communities of real women and men. Your task is to prototype an infrastructure of an interactive globe, where viewers can fluidly zoom in to connect to people's "best self", using their own verified images to represent their streets, homes and families. You’ll have the opportunity to share ideas with a leading game developer, on how to make the world fairer and more peaceful.