Artificial Intelligence Engineering (Intelligent Biomedical Systems Engineering), MS postgraduate personal statement example

Personal statement example

Halfway through testing my final-year project, I noticed that the heart rate my algorithm reported for one volunteer was rising and falling in time with the metronome rather than with her pulse. She was stepping on and off a low platform at a fixed rhythm, and the strongest frequency in the wrist sensor's optical signal was her arm swing, not her heartbeat. That plot changed the direction of my project. It is also why I now want to study how intelligent systems can be made reliable when they work with signals from real bodies.

The project, part of my BEng in Electronic Engineering, used a low-cost photoplethysmography sensor and a three-axis accelerometer mounted on a wristband I built from a development board. My first approach picked the largest spectral peak in a sliding window, and it worked well at rest. Once people moved, it failed in the way I have described. I then used the accelerometer as a noise reference. I implemented a least mean squares adaptive filter, added a simple rule that tracked the heart-rate peak from one window to the next, and compared both against a chest strap. Eight course-mates volunteered for short sessions of sitting, walking and stepping. The combined method reduced the error substantially during walking. It still struggled whenever cadence and pulse frequencies overlapped, and I reported that honestly rather than hiding it.

The most useful thing I learnt concerned evaluation. I had initially tuned parameters using all eight recordings, which flattered the results. Switching to leave-one-subject-out testing made the numbers worse but meaningful, because a wearable device will always meet a wrist it has never seen. Volunteers with darker skin tones or a looser strap fit also produced weaker signals. With eight people I could not draw conclusions from this, but it made me realise that a biomedical dataset is never a neutral sample. I would like to learn how to design data collection and validation with this in mind from the start.

Alongside the degree, and since graduating, I have worked at a bicycle repair shop. Most of the job is ordinary work: truing wheels, replacing brake pads and explaining to customers why a cheap chain wears out a cassette. It has given me a practical respect for diagnosis. A creak can come from the pedals, the bottom bracket or the seat post. Working through causes in order, and changing one thing at a time, is the same discipline I needed when debugging my filter. I have also become the person colleagues ask to set up customers' cycling computers and sensors, which has shown me how often the people who own these devices do not understand their readings.

Outside work I play tenor horn in a community brass band and look after its music library. That means cataloguing several hundred arrangements in a spreadsheet, tracking loans and making sure parts are ready before rehearsals. It is not glamorous, but it has made me organised and patient with records that other people rely on.

My technical preparation is strongest in signal processing, embedded programming in C and data analysis in Python. During my degree I took an introductory machine learning course and have since worked through further material on my own, including building simple classifiers for activity recognition on a public accelerometer dataset. I am aware that my knowledge of physiology is limited to what my project required, and I want postgraduate study to give me a firmer foundation in biomedical systems as well as in modern learning methods.

In the longer term, I would like to work on medical and wellbeing devices where the algorithm is only as good as its evidence. I am particularly interested in methods that remain trustworthy when the signal is noisy or the user is unlike the training data. I want to develop the engineering judgement to know when a model should give an answer, and when it should say it cannot.