BSc Computer Science · Targeting AI/ML MSc for Sept 2026

I'm a final-year BSc Computer Science student at Kingston University, applying to AI/ML MSc programmes for September 2026 entry.

My CS degree is software-engineering heavy — no formal modules in linear algebra, multivariate calculus, or probability. Rather than route around that gap, I'm closing it directly. Each project below pairs a from-first-principles mathematical derivation with a NumPy implementation that matches the derivation step by step. No autograd. No PyTorch shortcuts. No imports of sklearn.MLPClassifier.

• 🔭 Currently building pure-NumPy implementations of MLPs, Transformers, and Diffusion models
• 🌱 Currently learning linear algebra, multivariate calculus, and probability through Imperial's Mathematics for ML and Strang's MIT 18.06
• ⚡ Final-year project: Travelyn AI — iOS commuter app with on-device unsupervised pattern learning
• 📫 Reach me at zaidan2440@gmail.com

Languages
ML & Math
Web & Mobile
Databases
Cloud
DevOps & Tools
Operating Systems
Hardware
Game & 3D
Creative

Decoder-only Transformer trained on Tiny Shakespeare in pure NumPy. Scaled dot-product attention, multi-head decomposition, causal masking, and sinusoidal position encodings — all derived in notes/attention.md, including the variance argument for the √dₖ scaling.

Python NumPy Attention From Scratch

Two-layer multi-layer perceptron on MNIST in pure NumPy. Forward pass, backpropagation, and mini-batch SGD implemented to match the chain-rule derivation in notes/backprop.md, including the softmax + cross-entropy gradient collapse to p − one_hot(y).

Python NumPy Backpropagation MNIST

Multinomial Naive Bayes from first principles on the UCI SMS Spam Collection. Bayes' theorem, log-likelihood in log-space, and Laplace smoothing derived from scratch; benchmarked against the scikit-learn baseline.

Python Probabilistic ML Text Classification

Denoising Diffusion Probabilistic Models on Fashion-MNIST. Currently working through the variational lower bound derivation before implementation begins.

Python Diffusion Models Generative AI

iOS commuter app integrating four live UK transport APIs (TfL, National Rail Darwin, BODS SIRI-VM, OpenStreetMap) on a Backend-for-Frontend architecture. On-device unsupervised clustering (300 m greedy nearest-neighbour, three-visit minimum, time-bucketed) learns each user's commute patterns without training data; a separate pipeline interpolates live vehicle positions between scheduled and actual times. Verified with an XCTest unit suite and a five-participant Nielsen-heuristics usability study.

Swift SwiftUI Supabase iOS