M A T T Y   B E L L
0
About Work Projects GhostKernel Spotify Skills Contact
Software Engineer • AI & Automation

Hi, I'm Matty Bell

I build software around AI, automation, and machine learning — mostly tools that solve awkward problems and end up being genuinely useful.

Scroll to explore
Matty Bell
Python Machine Learning LLMs Java C# C++ Cyber Security YOLOv8 .NET GhostKernel Desktop Automation Python Machine Learning LLMs Java C# C++ Cyber Security YOLOv8 .NET GhostKernel Desktop Automation
01

About Me

Matty Bell at Northumbria University
Northumbria University Graduate

I’m a Software Engineer who likes building things that are practical, fast, and a bit ambitious.

I studied Cyber Security and Computer Networks at Northumbria University, which gave me a solid grounding in networking, security, and how systems behave when things go wrong.

Most of my work now sits around Python, Java, C#, and C++, whether that means building automation tools, desktop utilities, or applications that need to be reliable without becoming overcomplicated.

0+ Languages
0+ Projects
0+ Years Exp

When I'm not coding

⚽ Man United Fan 🎬 Better Call Saul 🎥 Django Unchained 🇷🇺 Learning Russian
02

Current Work

Oak Engage Newcastle upon Tyne
Current role

Technical Support Engineer

Technical support across Oak’s employee app and intranet platform

I currently work at Oak Engage in Newcastle as a Technical Support Engineer. Day to day that means handling support tickets, reproducing bugs, helping customers work through issues, and making sure problems are clearly documented for the teams fixing them.

Day-to-day

Diagnosing issues, reproducing bugs, and giving customers clear next steps without dressing simple answers up in a load of jargon.

Platform support

Working across Oak Sites, notifications, messenger, media processing, mail delivery, and the other moving parts people rely on every day.

Customer journey

Helping customers from onboarding through to launch and ongoing support, including pointing them towards the right training and help-centre resources.

Collaboration

Working closely with internal teams so recurring issues are raised properly and customer feedback actually turns into product improvements.

03

Ghost Kernel

Windows desktop automation built around the UI automation tree

GhostKernel glass overlay interface

What Is GhostKernel?

GhostKernel is a desktop automation layer for Windows that works through the Windows UI Automation accessibility tree instead of screenshots or pixel matching. A plain-English command like "click Compose in Outlook" gets broken down into something the app can resolve and execute safely.

The main reason I built it this way was reliability. Using accessibility data makes it far less fragile than screen-scraping, and every action is logged so there’s a clear audit trail of what happened.

Intent Resolution Pipeline

Each command passes through a few layers before anything is clicked or typed:

💬 User Command "click Compose in Outlook"
L1 Regex Parser Deterministic pattern matching
Zero latency, zero API cost
L2 Embedding Classifier ~22MB sentence-transformer
Runs locally and stays lightweight
L3 LLM Normalizer GPT-4o-mini (optional)
Confidence-gated fallback
L4 Smart Brain Planner Multi-step decomposition
Adaptive confidence thresholds
L5 Visual Grounding Live UIA tree scan
Real-time element matching
Safe Execution Preflight → Highlight → Execute → Audit

Key Features

🔮

Plain-English Control

You can type something like "find email from Clare" and GhostKernel works out the steps needed to do it.

🛡️

Safety-First Design

Three safety profiles (strict/balanced/permissive). Blocks destructive keywords, sensitive windows, and requires typed CONFIRM for risky actions.

🔗

Hash-Chained Audit

Every action is logged in sequence, so it is easy to trace what happened and when.

🧠

Self-Training AI

The project can generate tougher test commands for itself, making it easier to spot weak areas and improve the local models over time.

🎨

Glass Overlay UI

A translucent WebView2 overlay that can snap to the edge of the screen, move around easily, and surface live status while it runs.

🔌

Plugin SDK

Drop .py files into plugins/ and they're live. Skill contracts with can_handle() + execute(). Three built-in packs included.

🧩

60+ Python Modules

Fully modular architecture: engine, brain, NLP, skills, planner graph, visual grounding, screen intelligence, MCP bridge, and more.

📦

Single EXE Build

Ships as a standalone GhostKernel.exe via PyInstaller. No Python install needed. System tray, global hotkey, persistent SQLite storage.

Example Commands

GhostKernel — Interactive Session
ghost >click Compose in Outlook
✓ Found "Compose" button in "Outlook" → Highlighting → Clicked
ghost >type "Meeting at 3pm" into Subject in Outlook
✓ Found "Subject" field → Typed "Meeting at 3pm"
ghost >find me the cheapest 3070ti for sale
🔍 Web retrieval → Ranked 5 results → Opening top match
ghost >open youtube
✓ Resolved via site map → Launching browser → youtube.com
0+Python Modules
0+Test Files
0Pipeline Layers
0Safety Profiles
04

On Spotify

A couple of playlists from my profile.

Profile

Matty ßell

A couple of public playlists I actually wanted on here.

Open Spotify Profile

AI Overclock Tuner

GPU monitoring with LLM-backed recommendations

The AI Overclock Tuner is a GPU monitoring tool that reads real-time metrics like clock speeds, temperatures, power draw, and hash rates, then passes that data into an LLM layer to suggest better overclock settings.

It also includes a live dashboard, historical logging, and trend tracking, so it is easier to see what is actually working instead of guessing and hoping for the best.

🖥️ GPU Hardware Clock, temp, power, fans
01 Metrics Collector Real-time polling via .NET
Structured telemetry snapshots
02 Historical Logger Time-series data storage
Performance trend analysis
03 LLM Reasoning Engine Metric analysis
Recommendation generation
Optimized Settings Hash rate ↑ Power ↓ Stability ✓

NetDash Pollen ID

Computer vision for pollen grain detection and identification

NetDash is an end-to-end computer vision pipeline for automated pollen grain identification. It combines YOLOv8 object detection with Hough Circle Transformation to tighten up localisation and improve classification accuracy.

The pipeline processes microscopy images through multiple stages — from raw image preprocessing to neural-network-based detection, geometric refinement, and final species classification. Useful for environmental monitoring, allergy forecasting, and forensic palynology.

🔬 Microscopy Image Raw pollen slide capture
01 Image Preprocessing Resize, normalize, denoise
Color space conversion
02 YOLOv8 Detection Neural network object detection
Bounding box localization
03 Hough Transform Circle detection refinement
Precise grain boundary mapping
04 Species Classifier Feature extraction
Grain type classification
📊 Identification Results Species, count, confidence scores
05

Skills & Expertise

Languages

Python
Java
C#
C++

AI / ML

LLMs & NLP
Computer Vision
YOLOv8
ML Algorithms

Frameworks

.NET
PyTorch
TensorFlow
Flask / Django

Security

Network Security
Ethical Hacking
Digital Forensics
Penetration Testing

Education & Background

2020 — 2024

BSc Cyber Security & Computer Networks

Northumbria University

Developed expertise in network security, ethical hacking, digital evidence analysis, and software engineering fundamentals.

2018 — 2020

A-Levels

Biochemistry & Statistical Maths

Built a solid understanding of scientific principles, analytical thinking, and mathematical foundations.

06

Let's Connect

If you want to talk about a project, a role, or anything I’ve built, feel free to get in touch.

For paid contract work, please reach out by email.