From Data Scraping to Model Optimisation

Fedor Chursin, a first-year Applied Data Science and Artificial Intelligence student, spent eight weeks on a computer vision project that analyses photos of meals to estimate their carbon footprint, track eating habits and recommend adjustments.

Across the ML lifecycle

The work covered data preparation, modelling, evaluation and deployment. Fedor assembled 5,000-7,500 meal images by combining provided datasets with his own Python scraper built on the DuckDuckGo API. After cleaning, he trained neural networks to identify food items and pick up on visual patterns - shape and colour in foods such as beef.

Getting it to work

The first model managed only 25% accuracy. With guidance from lecturers, better preprocessing and an optimised architecture, that rose to 85-90%. The final deliverable was a mobile application whose UI and UX were shaped by peer and lecturer feedback.

Beyond the technical side, the project put weight on ethics - fairness and mitigating bias in particular. Fedor is aiming at a career in AI engineering, with an eye on further study, possibly in medical AI.