skip to content

Department of Computer Science and Technology

Part IB project demo montage banner

Candidates for Part IB of the Computer Science Tripos undertake a group design project as part of their practical work in Lent term. Each group has been working with a client from our group of industry supporters and has been developing a product based on an initial design brief over the last seven weeks. We would like to express our thanks to the clients for their continued support to the students. 

This year the pandemic has brought significant challenges. One in particular is the end-of-course presentations, which cannot be in person. Instead each student group has produced a short video presentation of their work for other students, faculty and clients to vote for. You can see all the videos on this page.

Prize Categories

The Lab will be awarding prizes to groups for the following achievements:

  • Most Impressive Technical Achievement
  • Most Impressive Professional Achievement
  • Best Contribution to a Better Future

Full Youtube Presentation Playlist


Individual Group Briefs and Presentations

Alpha: Aerial Video Selfies

Client: Cedric Barreteau, IMC

Alexander Hood, Anoushka Mazumdar, Aleksander Misztal, Boyue Zhang, Krzysztof Druciarek, Peter Wild

A recent viral video of a guy singing the Fleetwood Mac song Dreams, while skateboarding down the road, was swiftly followed by a similar video from Mick Fleetwood himself. But isn't it a bit dangerous for 73 year-olds to be filming themselves like that? Your task is to design a drone control system that follows a planned camera path while maintaining focus on a moving person, taking good quality aerial close-up video as they skateboard, bicycle, or just walk. It would be particularly impressive to use information like bluetooth signals and computer vision to automatically recognise and follow even an unplanned path. The result would be great for sports coverage, and could even be the future of mobile Zoom.


Bravo: Augmented Room Dressing for Zoom

Client: Matt Johnson and Olly Powell, Frontier

Alba Navarro Rosales, Gediminas Lelesius, Ondrej Baranovic, Tianhao Dong, Xavier Bugg, Yu He

In these times of Covid-19, many of us are spending much more time alone and interacting with colleagues and peers via webcams in virtual meetings. In the spirit of light relief, we would like the client to produce a fun application which would capture the video feed from a webcam, augment it with entertaining elements such as models, images or animations (For example, one option may be to provide a virtual set of bookshelves in the background to allow the user to appear better read for an interview!). The application should be able to work with some captured 3d information about the scene in order to place the friendly or amusing augmentations to brighten up the room of the user, or even provide some interactive background elements. The output should then be supplied as a virtual camera feed which may be consumed by applications such as Zoom or Teams.

Charlie: Boosting Skills after COVID-19

Client: Sherry Coutu, Digital Boost

Auguste Balzaryte, Aidan Brocklebank, Christopher Goh, Hannah Lord, Mark Englander, Matthew Sirman

Digital Boost is an online learning platform to help people who work for small businesses and charities to be mentored in ‘all things digital’, to enable them to come out of the COVID-19 crisis stronger. A 46-part taxonomy of skills currently covers 12,000 courses from 6 different providers. Your challenge is to create an intelligent navigation and guidance tool that creates a unique journey for each person, using multiple sources of information to build a model of that person's needs and experience, and drawing on opportunities for mentoring, workshops, short courses and other resources as appropriate to their needs.

Delta: Cascading Galton Boards

Client: Luke Baxter, Boeing

Alistair Westmacott, Federico Stazi, Leo Laine, Matthew Keenan, Samuel Ferguson, William Hayter-Dalgliesh

Many people struggle to learn basic ideas of probability distributions and sampling, despite the fact that we see these around us all the time. The Galton Board (see is an elegant and intuitive mechanical simulation that makes the origin of Gaussian distributions very clear. Unfortunately physical Galton Boards are costly to build. Your task is to make a customisable animated simulation - but not just of a single distribution. Users should be able to chain multiple boards together, associating them with descriptive equations, creating a visual probabilistic programming language that implements composition of Gaussian kernels as a zoomable giant pinball machine that can help teach conditional probability and an intuitive derivation of Bayes theorem.

Echo: Clinical Nursing for Children

Client: Michael Malley, Royal College of Paediatrics and Child Health

Andru-Lucian Stefanescu, Amy Nichol, Brendan Coll, Mital Ashok, Samuel Johnson, Sean Carey

In wealthy countries, hospitals use complex clinical information systems to capture and analyse data such as heart rate, respiration, oxygen saturation, temperature etc. In countries like Myanmar, nurses working in child healthcare have no access to computers that could be used to monitor and chart the condition of multiple patients every 1-2 hours over the course of their treatment. Your goal is to design a mobile app that can be used to capture data, generate clinically relevant output and empower nurses to escalate patient care as appropriate. Challenges will include transferring data between shifts, protecting patient confidentiality, and potentially integrating with existing apps.

Foxtrot: Computing for Bird Colonies

Client: Steffen Oppel, RSPB

Erin Cox, George Willcox, Maximilian Kaufmann, Robert Ramsay, Rishabh Jain, Ryan Shao

The RSPB have many hours of acoustic data recorded from bird colonies, which is becoming a valuable resource for conservation work such as monitoring changes in colony size, or the relative composition of different species. Your task will be to create a convenient set of tools for machine learning experiments using these data. You will be free to make your own suggestions, but tools for segmentation and classification will be important starting points. The tools should be packaged in a way that is convenient and easy for RSPB staff without computer science training.

Golf: Consignment Tetris

Client: Francesco Ciriello, MathWorks

Ana-Maria Radu, Connor Redfern, Joseph O'Connor, Marcus September, Mohammed Miah, Vladimir Mirjanic

Short-run manufacturers can now supply arbitrary bulk parts for 3D printing and online delivery, but manually packing a wide range of 3D shapes is difficult for robots. Your task is to design a system that will take a random selection of 3D models (e.g. as might be printed from, and use twin robot arms to fit them together into delivery boxes. You will simulate the system's kinematic and dynamic performance using Simulink 3D animation & Simscape Multibody in conjunction with a high-fidelity robotic arm model from MATLAB’s Robotics System Toolbox.

Hotel: Crossing the Bubbles

Client: Arley Anderson, Ab Initio Software

Ammaar Patel, Harrison Pitcher, Mikolaj Stepinski, Shahnoor Kiani, Thomas Quinnell

We are all learning a lot about the dynamics of virus transmission, but often in ways that leaves us having to trust experts who don’t themselves understand why some social settings result in “super-spreading,” others are benign “bubbles”, and how these very different situations interact when people in a city move between them. Your task is to create an intuitive environment that models and visualises the dynamics of infection and transmission both within and between diverse settings such as schools, private homes, and workplaces. The key challenge is presenting controls that allow people without mathematical training to experiment with causal parameters and understand the often non-linear consequences over time. The visualisations should be delivered via a web browser, intuitively readable by primary school children, but based on real evidence and mathematical models of transmission and infection dynamics.

India: De-biasing the Employment Process

Client: John Pettigrew, Umbrella Analytics

Ananya Hari Narain, Ilja Caikovskis, Joseph Cheng, Milly Kumar, Rebecca Tyson, Tsveta Todorova

The past year has seen a significant rise in awareness of misogyny, racism and other discrimination in society and, in particular, in workplaces. In addition, there is growing awareness of how ‘algorithms’ can reinforce bias rather than remove it. Your task is to produce a system that can help businesses recruit more fairly, by removing biased language from their job adverts that would put off many candidates. Your system should allow users to upload the text for a job ad, to identify problems using natural-language processing and statistics, and to recommend changes to the user so that they can make iterative improvements. Ideally, your system would give each text an overall score as well as word-level feedback.

Juliet: Deliberative Social Media

Client: Anna McKeon, Traverse 

Dan-Stefan Damian, Ermeyas Girma, Jonathon Ackers, John Higgins, Krzysztof Rymski, Keshav Sivakumar

EngagementHQ is a software platform designed for public engagement and debate, which was used alongside Zoom for recent research during the #LockdownDebate project. But in recent times, governments are having to move faster and faster, while the large audiences on social media struggle to engage in relevant ways. Your goal is to create a media platform that is as engaging as TikTok, but also allows thoughtful deliberation when different generations and communities have to share their perspectives and make compromises on difficult questions, even while government policy might be changing on a daily basis.

Kilo: Galapagoan Machine Learning

Client: Sophia Cooke, Cambridge Conservation Initiative 

Arjun Tapasvi, Andor Vari-Kakas, Hamza Hussain, Joey Chen, Karan Sharma, Seewon Choi

The residents of the Galapagos Islands could be more involved in conservation research, leading to educational and business opportunities, through using their mobile phone to interact with research data. Your client has field recordings of birdsong collected from the Islands, and your task is to use this to train machine learning tools that can identify species and bootstrap locally distinctive citizen science activities. You will need to provide cloud facilities for data management and analysis, as well as phone-based client software that allows the users to contribute to research activities in ways that bring them positive educational and economic benefits.

Lima: Instrument Landing App

Client: Phil Marsden, Softwire 

Benjamin Andrew, Hong Wong, Jan Letovanec, Kwing Li, Max Johnson, Shuntian Liu

Landing a plane using the instrument landing system (ILS) requires regular practice to keep the necessary skills up to date. Unfortunately, it's expensive and inconvenient for amateur pilots to get ILS landing slots so they can practice. The goal of this project is to simulate ILS using the GPS capabilities of a mobile phone, so that pilots can practice their ILS skills even during a normal landing. The phone screen should simulate actual ILS instruments when the plane is descending. You may need to think about how to test this system without access to a plane and airport - perhaps by providing descent guidance to cyclists coming into town down Castle Hill?

Mike: Intelligent Tools for Coeliac Disease Diagnosis

Client: Julian Gilbey, Lyzeum Ltd

Anastasia Courtney, Daniel Gooding, Declan Shafi, Edward Cassidy, King Yeh, Thomas Christie

In principle, computer vision and machine learning methods can be used to recognise coeliac disease. “Coeliac disease” is a very common autoimmune condition triggered by eating gluten, and treatable by following a gluten-free diet. Currently, in order to diagnose coeliac disease, pathologists manually inspect biopsies of gut tissue. We would like to automate this diagnosis process. Unfortunately, there are some difficulties; for example, different microscopes make the colours of the images look different, different samples look very different from each other, the images are huge, and we are interested in both small-scale and large-scale features. This project involves working on part of this process to allow for much faster and more accurate diagnosis, making use of computer vision and machine learning methods with other tools (for example image editors).

November: Managing Agile Researchers

Client: Ben Searle, TechWolf

Edward Groom, Gabor Pituk, James Hogge, Kimberley Worrall, Maximilian McGuinness, Thomas Barker

Software companies use agile project management to ensure that a backlog of feature requests is implemented in a timely way. Universities are not so agile, but often employ people with a wide range of technical skills, including computer officers and research engineers. Your task is to use natural language processing methods to automatically collect and classify skill areas from GitHub and StackOverflow accounts associated with Cambridge staff, and match these against feature requirements in a backlog of requests such as "collect lecture feedback", "improve performance of video download on Android" and so on. It’s likely that web pages or research publications associated with the individuals will provide further natural language clues to relevant application areas.

Oscar: Mapping the Missing

Client: Adam Basill, Metropolitan Police

Agnieszka Niewiadomska, Benjamin Thomas, Cameron Griffiths, Gardar Ingvarsson, Megan Finch, Radzim Sendyka

When people are reported missing to the Metropolitan Police, and specialist search advice is required to try and find them and return them to safety, Police Search Advisors are assigned to the case. One of the search tools used to assist locating the missing is documented statistics from previous cases. However, the current data used is 8 years old and none of the data studied was from London. Police Search advisors cannot be familiar with all the thousands of cases in their database and London's individual demographics are likely to display behaviours and patterns unique to London. Your task is to use (an anonymised version of) their database to help visualise and guide search efforts, based on the characteristics of similar cases in the past. Note that this team will need to interact with data that is sensitive and may be distressing.

Papa: Online Loop Jam

Client: David Russell, The Fusion Works

Alistair O'Brien, Joshua Pollick, Jiaxin Wang, Karl Mose, Maximilian Webb, Thomas Burrows

While many activities in Cambridge have transferred reasonably smoothly to the online world, music has not at all. It’s not even possible to sing Happy Birthday properly via videoconference, let alone coordinate choirs or band rehearsals. The Web RTC standard does provide support for time synchronisation of asynchronous messages, so together with interfaces such as Web MIDI, it should be possible to get better-timed music performance. Of course latency doesn’t go away (we still have the speed of light to deal with) so music made online can naver be like playing in the same room. Instead, we need music that can be played in cycles, so that each player contributes content that sounds good to their own ears now, and still sounds good after a fixed loop interval (perhaps 4 bars, perhaps 12, or even the next song verse), when the same notes are distributed and mixed in to the jam that everybody hears slightly later.

Quebec: Speeding up Evidence Synthesis for Conservation

Client: Phil Martin, Department of Zoology 

Eliot Makin, Markus Walder, Olivia Hayward, Tatiana Sedelnikov, Zeyu Cao

A core activity of the Conservation Evidence team in the Department of Zoology is compiling and summarising research literature, to provide evidence for which management approaches are most effective for biodiversity conservation. However, this evidence is dispersed in many different scientific journals, in English and non-English language journals, and in non-academic ("grey") literature. The Group needs tools that can use natural language processing methods to constantly monitor publications across many different venues, initially classifying studies as relevant or not to different biological taxa, and then extracting key attributes such as geography, habitat, threats or interventions and organising these for thematic browsing or targeted queries. The group would also like to explore the possibility of using natural language processing to summarise study findings.

Romeo: The New Internet of Things

Client: Saaras Mehan, Microsoft Research 

Alessandro Farace di Villaforesta, Henry Caushi, Marta Walentynowicz, Martynas Sinkievic, Navindu Leelarathna, Viktoria Csizmadia

Lives and businesses have been transformed by the pandemic, with network technology now more critical than ever. What are the implications for smart buildings and the Internet of Things? Which classic IoT use cases are now irrelevant, and what new opportunities are there? Your task is to use the Microsoft Azure Sphere IoT platform to address one of the new problems of the COVID-19 era. Your client can provide a range of accessories (e.g cameras, sensors and actuators), and your solutions should be documented openly (e.g. on GitHub) to enable broader benefits.

Sierra: Virtual Agronomist

Client: Charles Gentry, NIAB

Aravind Prabhakaran, Raphael Levy, Samuel Beer, Tautvydas Jasiunas, William Ashton, Xiaochen Tan

Each year NIAB publishes its Agronomy Strategy to help farmers grow successful crops. The comprehensive advice is based on NIAB's internal research and covers a range of topics including: crop choice and rotation, seed rate and establishment, weed control and fungicide strategies. NIAB would like to use this guide to produce a 'virtual agronomist', an expert system which can be used by farmers to answer their questions. The system should be able to accept a basic natural language question, e.g. "What seed rate should I use for late sown wheat?" and either show the relevant answer, or prompt for more information with further questions, e.g. "What variety of wheat have you sown?", and so on. The virtual agronomist should be accessed through either a website or messenger platform (e.g. WhatsApp). Basic reporting should be made available to administrators, such as frequent query subjects, to identify where to direct new research.

Tango: West Cambridge Airfreight

Client: Francesco Ciriello, MathWorks

Antonia-Irina Boca, Edward Weatherley, Nikola Georgiev, Naunidh Dua, Oluwatobi Adelana, Thomas Patterson

You are tasked with the development of an autonomous drone delivery system for light-weight package transportation around the West Cambridge site. By using high-fidelity drone models from the new MathWorks UAV Toolbox, you will develop an autonomous ground control system that coordinates one (or more!) drones around the site. Consider how to incorporate path and task planning, motor and plant control, and sensing & perception for navigation and obstacle avoidance. Evaluation in the realistic UE4 simulation will include simulated sensor data from 3D LIDAR readings and simulated camera input.

Uniform: Zoom into Books

Client: Vladimir Vilde, DX Analytics

Ana Dolinar, Isaac Dixon, Mikel Bober, Peixuan Song, Victoria Adjei, Weronika Wiech

Art and travel books often have beautiful images, but it’s frustrating that you can’t pinch to zoom as you would with a phone, to see arbitrarily high resolution details. The purpose of this project is to identify those times when a picture in a book or magazine corresponds to an existing high resolution image that is available online. Your Android app should work in augmented reality style, starting with a view of the book through the phone camera, but then seamlessly zooming by substituting high-resolution online data.