→ portfolio · → linkedin · → cv
$ ./whoami --verbose
name : Al Muqshith Shifan
role : MSc Computer Science @ Ontario Tech
tagline : network engineer by trade, dev by obsession, researcher by 2am
domain : where deep RL, programmable networks, and game dev collide
status : teaching agents to play games & networks to route themselvesRIDGE: State-Conditioned Reward Blending for Behavioral Coverage in Deep RL Game Agents Al Muqshith Shifan, Kevin Christopher Chua Accepted at IEEE Conference on Games (CoG) 2026 - Madrid, Spain (September 2026)
Conceptualized, implemented, written, and accepted in a single 20-hour overnight sprint during our first month of the MSc. RIDGE (Reactive Inter-persona Dynamic Goal Engine) uses one PPO agent that dynamically blends Explorer, Survivor, Craftsman, and Warrior persona reward weights via smooth sigmoid functions conditioned on internal game state, so a single agent shifts its play style mid-episode. With gratitude to Cristiano Politowski, Ali Neshati, and Dr. Loutfouz Zaman for their support and rapid feedback.
I like building things that play. My favorite bugs are the ones where the agent finds a strategy I never intended.
- RIDGE - one PPO agent, four personas (Explorer / Survivor / Craftsman / Warrior) blended with smooth sigmoids on the Crafter environment. Accepted at IEEE CoG 2026. It learns to survive by vibe.
-
- Local-AI-Dungeon - a fully offline text-adventure engine: a lightweight Python game loop wired into a local Gemma 3 / Ollama model. No cloud, no leash.
-
-
Re;Animate '26 - 1st place. Built a turn-based tactics game in AMOS BASIC on an Amiga, because constraints breed creativity.
-
I care about clean systems as much as clever ones - the kind you can read six months later without crying.
-
languages ########## python js c ml / rl ########.. pytorch gymnasium ppo networking #########. p4 cisco (ccna) ios-xe tooling #######... git docker linux ollama$ curl -s al-scripting.github.io/about-me # the full story lives here $ open ./links --up-top # portfolio / linkedin / cv
// still convinced the best software feels a little like a game.
-