Rivals of Aether II's take on Smash Bros. Classic Mode: fight escalating CPUs across stages up to a boss. I joined a small team to polish and ship Arcade Mode for launch, owning the post-match results screen that displays run stats and the stage rating.

With weeks left before launch, I pitched and prototyped a Break the Targets mode. In days I had the target entity and a prototype level editor for placing platforms, replacing the old workflow of hand-typing clockwise vertex values into an Unreal data class with no way to see them in the level. All 10 target levels shipped on time.

A two-week reverse engineering project to learn Unreal Engine, and ultimately what got me hired at Aether Studios. The original camera was strictly 2D and felt stiff coming from other 3D platform fighters. The goal: a Smash-like system that interpolates the camera in 3D so it feels intentionally directed rather than dragged around by the player.

A two-week reverse engineering project to learn Unreal Engine, and ultimately what got me hired at Aether Studios. The original camera was strictly 2D and felt stiff coming from other 3D platform fighters. The goal: a Smash-like system that interpolates the camera in 3D so it feels intentionally directed rather than dragged around by the player.

The system blends rotation, depth-axis translation, and depth-proportional zoom, adding rotation subtle enough to be nearly imperceptible but a big lift to camera feel. It computes a target 'interest' (the point to look at) from the bounding box of the camera subjects, then rotates slightly (5-10°) toward them to keep them in frame, tilting down near the ground and up near the blastzone. Position projects the camera's depth along the new rotation plus the interest.

The original prototype that inspired it is on my GitHub.

Training mode shipped barebones at launch, built alongside Arcade Mode and the target levels and heavily inspired by UnclePunch's Training Mode for Melee, my benchmark for what a training mode should be.

A later revamp brought it closer to Street Fighter 6: input history, input recording and playback, a much wider set of CPU behaviors, and a long list of community-requested quality of life features, all in about three to four weeks. The recording and playback system was the riskiest piece but came together fast once wired into the existing input pipeline, and it's the feature I'm most proud of from the project.

Beginner, intermediate, and advanced tutorial tiers, spanning nearly my entire contract. The beginner set came first and was the hardest to start: with no framework for guided lessons, much of the early work built a level scripting system I reused for every tutorial after.

The intermediate lessons (edgeguarding, ledge options, DI, tech chasing) were interrupted midway by the training mode rework, which I paused them to ship. By then the scripting system was mature enough to compose most of a lesson from existing pieces.

The advanced lessons were the hardest to land, where the line between "the player did it" and "the CPU made it look done" is thin, so success conditions had to be precise and went through many rounds of playtest revisions. Building all three tiers taught me to juggle long-running feature work while architecting a level system that scales to whatever lesson came next.

Eye Break Mode

A training mode I concepted and built, inspired by eggcercize from UnclePunch's Training Mode for Melee. Eyes appear around the stage and the player has to read each one and respond with the correct input before it disappears, turning reaction time and execution into something you can drill directly.

Edgeguard Mode

A training tool for practicing edgeguards: keeping a CPU character from recovering back to the stage. I concepted it months before it shipped. Once the last launch characters had their recovery logic, spawning a CPU offstage and letting it recover was quick to build. Prototyped in a few days, working within the week.