I've always been fascinated by the night sky, the idea that somewhere out there, thousands of other worlds are orbiting other suns. When I learned that NASA's Kepler telescope detected planets by watching stars flicker, I knew I had to try and build something that could do the same.

Each time a planet passes in front of its star, it dims the light ever so slightly, just a tiny blip in a sea of noise. I built a dual-branch 1D CNN in PyTorch that simultaneously studies the full orbital period and zooms in on the transit itself, fusing both perspectives to decide: real planet, or false alarm? The model catches 93.9% of all confirmed planet candidates, missing only 18 out of 360 in the test set.

Because missing a real planet matters far more than a false alarm, I applied cost-sensitive learning to the loss function, a decision that alone boosted recall by 12 percentage points. Final metrics: 97.7% AUROC, 93.9% recall on real NASA Kepler data.

I got tired of ads on every device in my apartment and wanted full ownership of my data, so I decided to build my own infrastructure from a $50 Raspberry Pi.

I set up a Pi-hole that now blocks about 30% of all network traffic. Then I added a 2TB NAS for personal file storage, learning the ins and outs of disk partitioning, ext4 journaling, and fstab along the way. Everything runs in Docker containers so nothing can break anything else.

Ever wondered what actually happens when you call fork()? I got to find out by building it from scratch. This was a project for my CPEN 331 Operating Systems class, where we extended an instructional operating system with real Unix functionality.

I implemented everything from file operations (open, read, write, dup2) to process management (fork, execv, getpid) and thread synchronization using mutex locks and condition variables.

University life can be expensive, so finding free food on campus is always a big help. Our team of 5 built a web app that scrapes Instagram posts from UBC clubs and event pages to find free food events in real-time.

We built the backend in Go and Node.js to handle the Instagram API integration, and wired up a TypeScript frontend with Google Calendar so you could add food events directly to your schedule.