Automate the boring stuff (getting round to reading this book (opens in a new tab)), and build interfaces around it all.
About Me
My usual workflow starts with reason, then the next few hours will be me fighting not to lose sight of that while it shapes up.
First off, though... who is Cole? Cole is a 22 year old Ugandan that's grown up with loads of nosy questions. How does a TV broadcast end up in your living room? How does a bank know what's in your account? Animations, games, the internet: nearly everything I loved led back to a computer, and I wanted in. I couldn't pick just one corner of it, so I picked computer science.
These days I'm in my third year at Uganda Christian University, majoring in AI and Robotics with data science alongside it (because you can't perfect automation without understanding data) and the plan is really simple: start each year as a completely different person at everything I do.
Experience
IT InternUganda Civil Aviation Authority, Entebbe, April to August 2026
Public Relations ManagerHonours College, UCU, November 2025 to November 2026
Class RepresentativeComputer Science, UCU
Web Designer and Manager (Intern)Makerere University Innovation Pod, February to September 2024
Achievements
Bronze HonourInternational Youth Math Challenge 2025
National mathematics camp invitewhere the country's top 100 students were tested to pick a team of 15
Africa's Talking hackathonwhere my team built NetDoctor
Education
BSc Computer ScienceUganda Christian University, AI and Robotics track. Current CGPA of 4.73 out of 5.
Cameras record far more than anyone can watch. I'm investigating whether computers can flag suspicious behaviour from body movement alone, and whether training on noisy real-world footage helps.
Captions fall apart in noisy rooms, and deaf and hard-of-hearing people pay the price. On a team of six, I'm helping build a system that reads lips alongside audio.
Intelligent systemsManuscript in preparation
A PRISMA-guided review of how computer vision detects suspicious activity in surveillance video. Yes, I read all the papers.
Intelligent systemsCourse paper
A philosophy paper on what happens when efficiency becomes an optimisation problem and the thing being optimised includes people.
Cameras record far more than anyone can watch. I'm investigating whether computers can flag suspicious behaviour from body movement alone, and whether training on noisy real-world footage helps.
Captions fall apart in noisy rooms, and deaf and hard-of-hearing people pay the price. On a team of six, I'm helping build a system that reads lips alongside audio.
Intelligent systemsManuscript in preparation
A PRISMA-guided review of how computer vision detects suspicious activity in surveillance video. Yes, I read all the papers.
Intelligent systemsCourse paper
A philosophy paper on what happens when efficiency becomes an optimisation problem and the thing being optimised includes people.
Cameras record far more than anyone can watch. I'm investigating whether computers can flag suspicious behaviour from body movement alone, and whether training on noisy real-world footage helps.
Captions fall apart in noisy rooms, and deaf and hard-of-hearing people pay the price. On a team of six, I'm helping build a system that reads lips alongside audio.
Intelligent systemsManuscript in preparation
A PRISMA-guided review of how computer vision detects suspicious activity in surveillance video. Yes, I read all the papers.
Intelligent systemsCourse paper
A philosophy paper on what happens when efficiency becomes an optimisation problem and the thing being optimised includes people.
Cameras record far more than anyone can watch. I'm investigating whether computers can flag suspicious behaviour from body movement alone, and whether training on noisy real-world footage helps.
Captions fall apart in noisy rooms, and deaf and hard-of-hearing people pay the price. On a team of six, I'm helping build a system that reads lips alongside audio.
Intelligent systemsManuscript in preparation
A PRISMA-guided review of how computer vision detects suspicious activity in surveillance video. Yes, I read all the papers.
Intelligent systemsCourse paper
A philosophy paper on what happens when efficiency becomes an optimisation problem and the thing being optimised includes people.
Cameras record far more than anyone can watch. I'm investigating whether computers can flag suspicious behaviour from body movement alone, and whether training on noisy real-world footage helps.
Captions fall apart in noisy rooms, and deaf and hard-of-hearing people pay the price. On a team of six, I'm helping build a system that reads lips alongside audio.
Intelligent systemsManuscript in preparation
A PRISMA-guided review of how computer vision detects suspicious activity in surveillance video. Yes, I read all the papers.
Intelligent systemsCourse paper
A philosophy paper on what happens when efficiency becomes an optimisation problem and the thing being optimised includes people.
I plan my days in a paper notebook. This tool turned 33 pages of those notes into a working Notion system that keeps running even when my computer is off.
On a team of two, I worked on predicting hourly air pollution in Ugandan cities from low-cost sensor data, then the model went live behind a dashboard.
A neural network that taught itself Snake from scratch, going from random moves to an average of 289 food on a 20×20 board.
Data and automationRunning
A small automation pipeline that gives this site an actual pulse.
Building and shaping systemsLive demo
An intern check-in app that swaps the paper sign-in sheet for one button, and shows supervisors who's in today.
Building and shaping systemsLive
A Spotify-style music player that streams real, full-length songs, built with plain JavaScript and no backend.
Building and shaping systemsInternship project
During my internship at the Uganda Civil Aviation Authority, I built a web-based system for handling passenger complaints, getting passengers, service providers and Authority staff into one workflow.
Building and shaping systemsInternship project
Same internship, different challenge: a simple Java desktop replica of an airport operational database, for practising how flight records are amended.
Building and shaping systemsHackathon prototype · Team project
A network diagnostics platform, built by a small team at the Africa's Talking hackathon, that helps customers and internet providers find exactly where a connection failed.
Intelligent systemsPrototype · Team project
A team project that uses computer vision to tell ripe tomatoes from unripe ones.
Building and shaping systemsIn development
A Linux desktop client that wants to get along with Google Quick Share.
Data and automationCompetition entries
Machine learning competition entries, from predicting financial stress using mobile money behaviour to predicting biodiversity using climate data.
I plan my days in a paper notebook. This tool turned 33 pages of those notes into a working Notion system that keeps running even when my computer is off.
On a team of two, I worked on predicting hourly air pollution in Ugandan cities from low-cost sensor data, then the model went live behind a dashboard.
A neural network that taught itself Snake from scratch, going from random moves to an average of 289 food on a 20×20 board.
Data and automationRunning
A small automation pipeline that gives this site an actual pulse.
Building and shaping systemsLive demo
An intern check-in app that swaps the paper sign-in sheet for one button, and shows supervisors who's in today.
Building and shaping systemsLive
A Spotify-style music player that streams real, full-length songs, built with plain JavaScript and no backend.
Building and shaping systemsInternship project
During my internship at the Uganda Civil Aviation Authority, I built a web-based system for handling passenger complaints, getting passengers, service providers and Authority staff into one workflow.
Building and shaping systemsInternship project
Same internship, different challenge: a simple Java desktop replica of an airport operational database, for practising how flight records are amended.
Building and shaping systemsHackathon prototype · Team project
A network diagnostics platform, built by a small team at the Africa's Talking hackathon, that helps customers and internet providers find exactly where a connection failed.
Intelligent systemsPrototype · Team project
A team project that uses computer vision to tell ripe tomatoes from unripe ones.
Building and shaping systemsIn development
A Linux desktop client that wants to get along with Google Quick Share.
Data and automationCompetition entries
Machine learning competition entries, from predicting financial stress using mobile money behaviour to predicting biodiversity using climate data.
I plan my days in a paper notebook. This tool turned 33 pages of those notes into a working Notion system that keeps running even when my computer is off.
On a team of two, I worked on predicting hourly air pollution in Ugandan cities from low-cost sensor data, then the model went live behind a dashboard.
A neural network that taught itself Snake from scratch, going from random moves to an average of 289 food on a 20×20 board.
Data and automationRunning
A small automation pipeline that gives this site an actual pulse.
Building and shaping systemsLive demo
An intern check-in app that swaps the paper sign-in sheet for one button, and shows supervisors who's in today.
Building and shaping systemsLive
A Spotify-style music player that streams real, full-length songs, built with plain JavaScript and no backend.
Building and shaping systemsInternship project
During my internship at the Uganda Civil Aviation Authority, I built a web-based system for handling passenger complaints, getting passengers, service providers and Authority staff into one workflow.
Building and shaping systemsInternship project
Same internship, different challenge: a simple Java desktop replica of an airport operational database, for practising how flight records are amended.
Building and shaping systemsHackathon prototype · Team project
A network diagnostics platform, built by a small team at the Africa's Talking hackathon, that helps customers and internet providers find exactly where a connection failed.
Intelligent systemsPrototype · Team project
A team project that uses computer vision to tell ripe tomatoes from unripe ones.
Building and shaping systemsIn development
A Linux desktop client that wants to get along with Google Quick Share.
Data and automationCompetition entries
Machine learning competition entries, from predicting financial stress using mobile money behaviour to predicting biodiversity using climate data.
I plan my days in a paper notebook. This tool turned 33 pages of those notes into a working Notion system that keeps running even when my computer is off.
On a team of two, I worked on predicting hourly air pollution in Ugandan cities from low-cost sensor data, then the model went live behind a dashboard.
A neural network that taught itself Snake from scratch, going from random moves to an average of 289 food on a 20×20 board.
Data and automationRunning
A small automation pipeline that gives this site an actual pulse.
Building and shaping systemsLive demo
An intern check-in app that swaps the paper sign-in sheet for one button, and shows supervisors who's in today.
Building and shaping systemsLive
A Spotify-style music player that streams real, full-length songs, built with plain JavaScript and no backend.
Building and shaping systemsInternship project
During my internship at the Uganda Civil Aviation Authority, I built a web-based system for handling passenger complaints, getting passengers, service providers and Authority staff into one workflow.
Building and shaping systemsInternship project
Same internship, different challenge: a simple Java desktop replica of an airport operational database, for practising how flight records are amended.
Building and shaping systemsHackathon prototype · Team project
A network diagnostics platform, built by a small team at the Africa's Talking hackathon, that helps customers and internet providers find exactly where a connection failed.
Intelligent systemsPrototype · Team project
A team project that uses computer vision to tell ripe tomatoes from unripe ones.
Building and shaping systemsIn development
A Linux desktop client that wants to get along with Google Quick Share.
Data and automationCompetition entries
Machine learning competition entries, from predicting financial stress using mobile money behaviour to predicting biodiversity using climate data.
I plan my days in a paper notebook. This tool turned 33 pages of those notes into a working Notion system that keeps running even when my computer is off.
On a team of two, I worked on predicting hourly air pollution in Ugandan cities from low-cost sensor data, then the model went live behind a dashboard.
A neural network that taught itself Snake from scratch, going from random moves to an average of 289 food on a 20×20 board.
Data and automationRunning
A small automation pipeline that gives this site an actual pulse.
Building and shaping systemsLive demo
An intern check-in app that swaps the paper sign-in sheet for one button, and shows supervisors who's in today.
Building and shaping systemsLive
A Spotify-style music player that streams real, full-length songs, built with plain JavaScript and no backend.
Building and shaping systemsInternship project
During my internship at the Uganda Civil Aviation Authority, I built a web-based system for handling passenger complaints, getting passengers, service providers and Authority staff into one workflow.
Building and shaping systemsInternship project
Same internship, different challenge: a simple Java desktop replica of an airport operational database, for practising how flight records are amended.
Building and shaping systemsHackathon prototype · Team project
A network diagnostics platform, built by a small team at the Africa's Talking hackathon, that helps customers and internet providers find exactly where a connection failed.
Intelligent systemsPrototype · Team project
A team project that uses computer vision to tell ripe tomatoes from unripe ones.
Building and shaping systemsIn development
A Linux desktop client that wants to get along with Google Quick Share.
Data and automationCompetition entries
Machine learning competition entries, from predicting financial stress using mobile money behaviour to predicting biodiversity using climate data.
I wanted to see how high a snake could score if nobody told it how to play, and it could only learn from its own games.
What I did
I split the game rules from the graphics so training could run without a window, then trained a Double DQN agent on them. It sees 21 numbers from the snake's point of view, such as which moves are fatal and how much of the board each move leaves reachable.
The first version peaked at 188, then got worse. Looking at how it died showed why: it would cut off its own tail, then die about 100 moves later, far too long after the mistake to learn from it. A good snake also dies so rarely that it barely had any failures to study. So I gave it a separate memory of the moves before each death, and made dying cost more.
Evidence
On 100 boards it had never seen, it averaged 289 and reached 343 at best. A perfect game is 397, and random moves average 0.1.
A simple rule I wrote by hand, using the same inputs, still averages 345, so the learned model doesn't beat it yet. That also shows how much of the snake's skill comes from the inputs I designed. The final model rarely traps itself now, but it has become so careful that it tends to starve instead, which is the next thing to fix.