The UROP is designed to support undergraduates studying at the University of Cambridge who are going to return for at least one more year of undergraduate study.

Final year undergraduates and postgraduate students should not apply.

Some projects with external funding have additional restrictions, such as those funded by EPSRC.

If you have any questions please contact Joe Goddard, Industrial Placements Coordinator, who administers UROP projects for the Department of Engineering.

Further information can be found below:

• Student click here.
• Staff click here.

Available Projects

---

RTL/Verilog Communications Bus

Primary Supervisor Details

Name: Jon Bonsor-Matthews
Department: Electrical Engineering
Email Address: jpm66@cam.ac.uk

Co-Supervisors / Industrial Collaborators 

Name: Matthew Tang
Department/Company: Electrical Engineering
Email Address: wct26@cam.ac.uk

Project Description 

We are developing a new RTL (Verilog) lab for IIA to standardise, implement and test a new and custom communication bus. This UROP project is to investigate what can be achieved on the FPGA platform that we can use and to provide a starting point for the IIA students.

The project objectives are:

• Investigate existing open-source communication bus implementations
• Demonstrate standalone testing of that implementation
• Integrate the implementation on an FPGA with a processor
• Investigate existing drivers for the implementation
• Write a firmware test framework to test the communication bus in a loop-back mode
• Test the same across two FPGAs
• Prepare these findings in a clearly documented way, to allow for preparation of the IIA lab

Additional Information

This is a great opportunity for you to get your hands dirty writing and testing Verilog code, seeing how RTL blocks interface with a processor and how firmware drivers and application code is used with such an architecture.

Essential Knowledge, Skills, and Attributes

You should feel that you’d be able to come up to speed with Verilog and C very quickly.  Being able to show why you are interested in these languages is important.

You should be well organised, able to work independently with support and able to produce clear, concise documentation.

Timing

• Application opening and closing dates: Now - Weds 18^th March 2026
• Project start and end dates: 17^th Aug - 25^th Sep 2026

Continuation Opportunities

This UROP could lead into a 4^th-year project to continue research into digital communication buses.

Application Details

Please email Jon Bonsor-Matthews, jpm66@cam.ac.uk, with a copy of your CV along with a short statement in your email explaining why you are interested in this particular project. 

Deadline for applications: Weds 18^th March 2026.

Funding Status of your UROP

Fully funded (Div. B. & Teaching Office).

---

Beyond Accuracy: Confidence-Aware Speech Models for Alzheimer’s Disease Detection

Primary Supervisor Details

Dr Mengjie Qian (mq227@cam.ac.uk)

Department of Engineering

Co-Supervisors / Industrial Collaborators 

Dr Petar Raykov (Petar.Raykov@mrc-cbu.cam.ac.uk)

Memory and Aging Research Group, MRC Cognition and Brain Sciences Unit, University of Cambridge

Prof Kate Knill (kmk1001@cam.ac.uk)

Department of Engineering

Project Description 

Figure 1: Typical progression of AD compared to normal cognitive ageing with potential target classification boundary.

Alzheimer’s Disease (AD) is the leading cause of dementia worldwide, with global costs exceeding one trillion US dollars and expected to double by 2030. Early detection is crucial for timely intervention. Spontaneous speech offers a non-invasive window into cognitive function: changes in linguistic structure, lexical diversity, and prosody appear long before symptoms become severe. Advances in speech technology allow these markers to be analysed automatically, positioning speech as an accessible biomarker for cognitive decline.

Most AD detection studies focus on accuracy. State-of-the-art systems combine acoustic features, automatic speech recognition (ASR) transcripts, and linguistic embeddings to classify AD and healthy control speech, achieving 80-95% accuracy on datasets such as ADReSS and Pitt. ASR quality has also been explored. A key limitation remains: current systems produce predictions without confidence estimates, i.e. the systems cannot indicate how reliable a decision is. For clinical use, knowing when a model may be wrong is as important as accuracy. Overconfident errors can mislead decisions, whereas confidence-aware models support safer, more interpretable outputs.

This project addresses that gap by developing and evaluating confidence-aware speech models for AD detection. The work will be carried out by a UROP student over 10-12 weeks, with funding supporting up to 12 weeks. The project is organised into the tasks below:

1. Data preparation

The student will primarily work with publicly available datasets (e.g. ADReSS, ADReSSo, Pitt Corpus). The student will also review internal resources from the MRC Cognitive and Brain Sciences Unit (CBU) such as conversational or clinical recordings to document what data exists and what would be needed for future use.

2. Baseline model development

The student will implement a baseline AD model using combined acoustic and text embeddings based on our prior work [1], with accuracy, F1, and AUROC metrics as benchmarks.

3. Confidence estimation and calibration

This phase integrates confidence estimation into the baseline system. Methods may include softmax scores or a lightweight uncertainty module. Post-hoc calibration methods, such as temperature scaling and Dirichlet calibration, will align predicted confidences with true correctness likelihoods. Calibration will be evaluated using Expected Calibration Error (ECE), Maximum Calibration Error (MCE) and Brier score. The student will also assess how well calibrated models support reliable decision-making by allowing the model to withhold low-confidence predictions, analysing the trade-off between accuracy and coverage.

4. Confidence-based Interpretability (optional, depending on time)

The student will explore whether confidence scores can enhance interpretability, such as by identifying segments or features linked to high confidence. Methods may include token-level confidence visualisation or saliency/attention analyses to highlight confidence speech segments.

5. Documentation and reporting

This includes organising the codebase, adding documentation and configuration files for experiments and data use, and producing a project report summarising methods, findings, and recommendations for future work.

[1] K. E. Jackson. Alzheimer’s Dementia Recognition Using Speech Foundation Models. University of Cambridge, 2025.

Additional Information

This UROP will suit a student who is interested in Information Engineering and has taken modules such as 3F1, 3F3 and 3F8 in IIA, and is interested in taking courses such as 4F10 and 4F13 in IIB

Essential: Good python coding skills.

Desirable skills: pytorch, cuda, experience with deep learning models, Linux OS, strong Maths.

Timing

We are open to the UROP being for 10-12 weeks depending on the selected student’s preference.

• Application closing dates: 27th February 2026.
• Application interviews: week of 9th March 2026.
• Project start time: 29th June 2026.
• Project end time: 4th September 2026 (10 weeks) or 18th September 2026 (12 weeks).

Continuation Opportunities

This project could serve as a basis to a follow-on 4th-year project.

Supporting Information

Dr Mengjie Qian's Website: https://mi.eng.cam.ac.uk/~mq227/

Poster of previous related work: https://bit.ly/4aRq1Nc

Cambridge Language Sciences: www.languagesciences.cam.ac.uk/incubator/">https://www.languagesciences.cam.ac.uk/incubator/

Application Details

Please email Dr Mengjie Qian, mq227@cam.ac.uk, with a copy of your CV along with a short statement in your email explaining why you are interested in this particular project.

Deadline for applications: 27th February 2026.

Funding Status of UROP

Fully funded (Cambridge Language Sciences Incubator Fund Award).