I'm Emmanuel Mensah, a first-year Computer Science (Software Engineering) student at Brunel University London. My work sits at the intersection of desktop application development, AI agent systems, and creative tooling — I like building things that solve problems I actually run into, not just projects for a CV.
Before Brunel I studied at Newham Sixth Form College (NewVIc), where I completed a Level 3 BTEC in Computer Science and Games Development with triple-distinction grades and served as a Student Representative on the Student Council. That's where I started taking software seriously , games coursework pushed me into OOP and mobile development, and I kept building from there.
Outside of coursework, a lot of my learning comes from hobbies colliding with engineering. I edit videos in After Effects, and when I got tired of paying for upscalers that were slow or bad, I built my own in C#/WPF using Real-ESRGAN, which meant learning FFmpeg pipelines and patching MP4 metadata at the byte level because standard libraries couldn't handle my edge cases. I still use it on my own edits. That's the kind of work I enjoy most: picking a real problem and going deep enough to actually fix it.
More recently I've been building multi-agent AI systems. For the Google Gemini Live Agent Challenge 2026, I built a system that autonomously operates a Linux desktop to perform engineering design tasks — running CAD automation in FreeCAD, web research through Playwright with Gemini vision, and a planner module powered by Gemini 3.1 Pro that routes requests across specialist agents. Getting the agents to self-correct from errors instead of crashing was the hardest and most rewarding part.
I'm actively expanding into full-stack web development (React + FastAPI) alongside sharpening my core CS fundamentals. If you're working on anything interesting in AI agents, desktop tooling, or creative software — reach out.
Object-Oriented Programming · Multi-Agent Systems · Computer Vision · Desktop Application Development · Video Processing · API Design · Event-Driven Architecture · Full-Stack Web Development · REST APIs · CI/CD
NB: This is a living document reflecting what I'm actively using and what I'm learning. Some tools are part of my daily workflow, others I'm still building fluency in. Growth is the point.
Python · Google Gemini API · Playwright · FreeCAD · OpenCV
A multi-agent system that autonomously operates a Linux desktop to perform engineering design tasks using Google Gemini's Computer Use API for vision-based screen control.
- CAD Agent — designs 3D parts in FreeCAD through menu-driven GUI interaction and Python macro execution, with dynamic error capture and self-correction
- Research Agent — uses Playwright and Gemini vision to search the web and extract structured data with confidence scores
- Documentation Agent — takes research output and generates automated reports
- Planner Module — powered by Gemini 3.1 Pro, routes complex requests across agents and generates step-by-step FreeCAD workflows
C# · WPF · .NET · FFmpeg · Real-ESRGAN
A Windows desktop tool that upscales videos using Real-ESRGAN AI models. Built because I edit video in After Effects as a hobby and existing upscalers were either too slow, too expensive, or both — so I built my own.
- Full pipeline from frame extraction to re-encoding with FFmpeg
- Custom binary patcher that fixes MP4 timescale and duration metadata directly at the byte level
- Supports both 32-bit and 64-bit atom formats
Adobe After Effects · ExtendScript/JavaScript
An After Effects script that automates common composition workflows, cutting down repetitive manual work in my editing pipeline.
- Panel-based UI for streamlined access to frequently used animation and editing operations
- Lets users quickly fill in and manage layer properties across compositions
Java · SwiftBot API · Raspberry Pi
An interactive Simon Says-style memory game built for the SwiftBot robot.
- LED colour sequences, button input, difficulty progression, and a lives system
- Individual underlight control mapping colours to specific LED positions
- Celebratory robot dance animations triggered on high scores
Java · SwiftBot API · QR Code Input
An OOP-based zigzag navigation program for the SwiftBot robot.
- Reads length and section count from QR codes, validates input, executes precise zigzag movements with 90-degree turns
- Path retracing using a stack data structure
- LED colour feedback for odd/even sections and full execution logging to file
More projects on my GitHub — thanks for stopping by.



