I build end-to-end machine learning systems with a focus on healthcare and clinical applications. My background spans deep learning, computer vision, NLP, EEG signal processing, and applied data engineering, with hands-on experience in NHS inpatient settings supporting young people with neurodevelopmental conditions.

Predicts Premier League match outcomes (Home Win / Draw / Away Win) with calibrated confidence scores, using an XGBoost classifier trained on 2014–2025 historical data. Features include rolling form, rest days, head-to-head history, and Elo ratings, all engineered with strict guards against data leakage. Reached ~51.5% accuracy on a chronological hold-out set (vs. 42.3% for the majority baseline), with a Streamlit app for browsing fixtures and tracking real-world prediction accuracy over time. Built as a personal ML engineering exercise, with an honest account of its limitations (draw prediction, no injury data) documented in the repo. (Private repository)

Automated sleep staging from raw EEG signals using a 1D CNN, MNE-Python, and interpretable AI. Built on the Sleep-EDF Expanded dataset (22,683 epochs, 13 subjects). Achieved balanced accuracy of 0.718 and Cohen's Kappa of 0.430, a large improvement over a Random Forest baseline. Applied SHAP and Grad-CAM to explain model decisions — the CNN independently identified sleep spindle locations without explicit labelling. Clinically motivated by sleep difficulties in children with autism, ADHD, and epilepsy. → github.com/danielakbank/eeg-sleep-staging

Interactive GIS project mapping real M1 and M6 motorway junctions, built around live junction data fetched from OpenStreetMap via the Overpass API. Builds maintenance priority zones with proper coordinate reprojection (WGS84 to British National Grid) and runs a spatial join to flag simulated roadside assets falling within those zones. Demonstrates practical spatial analysis with GeoPandas, Shapely, and Folium, deployed as an interactive Streamlit app. → Live app · github.com/danielakbank/uk-motorway-asset-mapper

Search tool for finding tradesmen (plumbers, electricians, carpenters, etc.) near any location, built on the Google Places API (New). Returns contact details, ratings, and Google Maps links, with cached searches and session rate limiting to control API costs. → github.com/danielakbank/tradesman-finder

U-Net with ResNet50V2 backbone for lesion detection and segmentation. Validation Dice ~0.71. Two-phase transfer learning. Deployed on Hugging Face Spaces via Gradio. → Live demo · github.com/danielakbank/BUSI-Segmentation

Compared classical CV preprocessing against a U-Net with EfficientNetB0 encoder on the UIEB benchmark. PSNR improvement of +3.03dB over the raw baseline, using two-phase transfer learning and a combined MAE/SSIM loss. Includes an interactive Gradio demo app. → github.com/danielakbank/underwater-image-enhancement

AI-powered job matching platform aggregating live listings from Adzuna, Reed, and Remotive APIs, scored against uploaded CVs using an LLM-driven engine. Deployed on Streamlit Cloud. → github.com/danielakbank/cv-job-matcher

Python ETL pipeline ingesting 5M+ DfE records across 150+ UK local authorities. Engineered a custom risk-flag combining re-referral rates and case duration, surfaced through an interactive Power BI dashboard built for non-technical stakeholders. → github.com/danielakbank/childrens-social-care-analysis

Streamlit app that turns any uploaded CSV into automated visualisations, statistical summaries, and plain-English AI insights, powered by Mistral running locally via Ollama for private, cost-free analysis. → Live demo · github.com/danielakbank/Data-Insight-Generator

End-to-end NLP app classifying product reviews as positive or negative, trained on 500,000+ Amazon Fine Food reviews. Logistic Regression matched a neural network at 94% accuracy while being faster and lighter to deploy. Includes cross-category breakdowns, sentiment trends over time, and keyword extraction. → Live demo · github.com/danielakbank/review-sentiment-analyser

Breathing exercise app built with Flutter. Published on the Google Play Store.

ML & AI

1D CNN · U-Net · Transfer Learning · NLP (TF-IDF) · LLM Integration (Mistral/Ollama) · EEG Signal Processing (MNE-Python)

Data Engineering

ETL Pipelines · API Integration · Web Scraping

Geospatial

Shapely · Coordinate Reference Systems · Spatial Joins

Visualisation & BI

DAX

Software & Deployment

Hugging Face Spaces

• Healthcare Support Worker, Cygnet Health Care CAMHS (2025–present)
• MSc AI & Data Science, University of Hull (2023–2024)
• BSc Computer Science, ESAE University (2018–2021)