Gameplay engineer. I build server-authoritative multiplayer in Unreal Engine 5.

Slammer is a multiplayer FPS in open playtest on Steam. Hide Seek Survive is a first-person horror game released in 2025. Both built alone: netcode, AI, weapons, UI, and everything else.

Slammer

Multiplayer FPS in open playtest on Steam. Windows, Linux, and Steam Deck. Built solo.

32 weapons. 24 attachments. Zero hardcoded values.

Every weapon in Slammer runs on the same system. Twelve components handle fire, ballistics, recoil, ADS, sway, aim assist, networking, and more, all communicating through a central mediator. None of them replicate individually. The weapon actor replicates a weapon ID and an attachment array, and each client reconstructs the rest locally.

The whole pipeline is driven by DataTables covering stats, cosmetics, attachment stats, and attachment cosmetics. Adding a weapon means adding a row, not writing code. All 32 weapons across 7 categories shipped that way.

Client-side feel. Server-side authority.

Each shot checks travel time to the target. If it’s fast enough, it resolves as hitscan. If not, it spawns a simulated projectile. Both paths share the same damage pipeline.

The client fires optimistically for responsiveness, but the server validates every hit. It checks timestamp, fire direction, and hit claim angle. If any check fails, the hit is rejected. Combined with server-side lag compensation verified up to 200ms, players get responsive gunplay without giving up authority. The host acts as the server, and the architecture is unchanged for dedicated.

Bots that play by player rules.

The bots call the same input functions a human does. No privileged state access, no aim assistance, no reading positions through walls. If a bot knows where you are, it saw you.

A role-based behavior tree handles per-role target selection and self-care behaviors like breaking off to seek cover and plate up. An EQS-driven cover system is in active development.

Hide Seek Survive

First-person horror game, released on Steam. Rebuilt and re-released in 2025. Built solo.

An AI that adapts to how you play.

The demon has two modes. In Seek, it camps patrol points 80% of the time, waiting to ambush. In Survive, triggered when the player finds objectives too quickly, it turns erratic: camping only 20% of the time and reversing direction unpredictably. The switch is driven by player progress, not a timer, so skilled players face a harder and weirder demon.

The system I’m most proud of is invisible. Early versions had a rage-quit problem. Jumpscares happened too often and players left. The fix was a cooldown that silently teleports the demon to the farthest patrol point when the player has seen a scare too recently and is getting close to where it’s hiding. No audio cue, no visual tell. The game feels dangerous the entire time, but the pacing is controlled behind the scenes.

The house is the other enemy.

Every room has four variants that load via level streaming. When the player exits and closes a door, there’s a 40% chance the room swaps to the next variant in a pre-shuffled sequence. The player’s room, the demon’s room, and the room with the active objective are protected from shuffling. The navmesh updates dynamically so the demon’s pathing always respects the new layout.

Every search is a risk.

The goal is to find the doll seven times to trigger the endgame. It hides under furniture, so players switch into a constrained drag mode built as a custom movement mode in the Character Movement Component, locking input to a 1D track along the furniture’s axis. Each draggable object carries a nav modifier, so the navmesh updates as furniture moves and the demon’s pathing reflects the current state of the room. The doll respawns after each discovery on escalating timers, giving the demon more time to hunt between finds.

Professional Experience

PROFESSIONAL EXPERIENCE

Booz Allen Hamilton — Lead Gameplay Engineer

I lead UE5 engineering for simulation and digital twin programs. Real-time crowd simulation scaling 300+ NPCs with navmesh-validated spawning and behavior tree driven stuck detection. An interface-driven save and load system persisting furniture, camera, and lighting independently across 2,000+ runtime-spawned actors. Before that, VR marksmanship training modules holding 60 FPS on Quest 3.

Same engine, same language, different problem space.

Contact

Open to gameplay and AI engineering roles. Based in Pennsylvania but open to relocation.

steverickley1@gmail.com