Skip to content
View zurielsingh's full-sized avatar

Block or report zurielsingh

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
zurielsingh/README.md
Typing SVG

🧠 About Me

I'm a Computer Science graduate with a focus on Machine Learning, Computer Vision, and AI — currently completing my Honours with a research project on sequence modelling for clinical disease progression.

What drives my work:

  • 🔬 Rigour — I care about leakage prevention, validation methodology, and reproducibility as much as the model itself
  • 🛠️ End-to-end builds — from system architecture through to evaluation and writeup
  • 🧩 Different domains — deep learning, NLP, multi-agent LLM systems, game dev, application dev
  • 📚 Currently exploring State Space Models (Mamba) and Transformer architectures for healthcare

🛠️ Tech Stack

Languages


Python

C++

C#

Java

JavaScript

SQL

HTML

CSS

Machine Learning & Data


PyTorch

TensorFlow

scikit-learn

Hugging Face

Pandas

NumPy

Game Dev & Graphics


Unity

SFML

Shader Graph

Tools


Git

GitHub

Linux

VS Code

Bash

🚀 Featured Projects

Next-visit MMSE prediction using gated recurrent neural networks on the OASIS-2 longitudinal dataset.

Highlights:

  • BiLSTM achieves R² = 0.75 on held-out test
  • 108-configuration hyperparameter sweep
  • Strict patient-wise leakage prevention
  • 5-fold GroupKFold cross-validation

Python TensorFlow Keras Sequence Modelling

A multi-agent system that autonomously generates, validates, and executes Python code.

Highlights:

  • Mistral-Instruct + DeepSeek-Coder orchestration
  • Mem0-based persistent memory
  • Sandboxed execution environment
  • Real-time documentation retrieval

Python LLMs Multi-Agent Mem0

Transformer-based résumé scoring system using BERT regression for consistent 0–100 quality scores.

Highlights:

  • BERT-based regression model
  • Achieved R² = 0.993, MAE = 1.38
  • Validation-based checkpointing
  • Substantial improvement over baseline

Python BERT HuggingFace NLP

3D third-person adventure platformer in Unity centred on a toy robot's journey to rebuild himself.

Highlights:

  • AI gameplay via FSMs and Q-Lite RL
  • 6 custom shaders (Toon, Glow, Hologram, etc.)
  • Modular dialogue and physics systems
  • Built with Shader Graph + HLSL

Unity C# HLSL Shader Graph

A C++ narrative-driven game with branching storylines and multiple endings driven by player decisions.

Highlights:

  • Built with the SFML framework
  • Modular OOP architecture
  • STL-based game state management
  • Multiple endings via decision trees

C++ SFML OOP Game Dev

🔬 Honours Research Project (In Progress)

Comparative study of Mamba-based State Space Models vs Transformer architectures for disease progression modelling on longitudinal clinical data.

Focus areas:

  • Long-sequence robustness
  • Irregular time-gap handling
  • Computational scalability
  • AUROC, AUPRC, Concordance Index

PyTorch Mamba Transformers Healthcare ML


🌱 Currently Learning

  • Computer vision & image processing — beyond classification, into spatial reasoning and segmentation
  • State space models — Mamba, structured state evolution, linear-time sequence modelling
  • Optimisation & modelling — mathematical techniques for structured decision problems
  • Healthcare ML — longitudinal data, irregular sampling, time-to-event modelling
  • Ontologies & knowledge representation — structured reasoning and knowledge bases
  • Agentic AI — multi-model orchestration, tool use, persistent memory

📫 Get in Touch


Pinned Loading

  1. dementia-progression-forecasting dementia-progression-forecasting Public

    Next-visit MMSE prediction using gated recurrent neural networks on the OASIS-2 longitudinal dataset — BiLSTM achieves R²=0.75 across a 108-config hyperparameter sweep with strict patient-wise leak…

    Jupyter Notebook

  2. fractured-in-time fractured-in-time Public

    Forked from UKZN-COMP315-Group15/fractured-in-time

    A C++ choice-driven quiz adventure game with branching storylines

    C++

  3. Python-Code-Assistant Python-Code-Assistant Public

    Forked from UKZN-GroupProjs/Python-Code-Assistant

    A multi-agent python application that assists users in programming in Python

    Python

  4. Restore-Rusty Restore-Rusty Public

    Forked from UKZN-GroupProjs/Restore-Rusty-Final

    A 3D Unity game developed as part of the COMP307 currirculum for the University of KwaZulu-Natal. Please note: we do not claim the use of any copyrighted media used in this project. All assets are …

    C#

  5. Resume-Evaluator Resume-Evaluator Public

    Transformer-based résumé scoring system using BERT regression to generate consistent, objective 0–100 résumé quality scores.

    Jupyter Notebook 1