B.IT Networking & IT Security · MSc Computer Science · Ontario Tech

ソフトウェア設計 I Do software design, networking, cybersecurity, game development, and AI research.

Jack of all trades, master of none, but better than most. Networking and security by training, software design and AI by research. I use games as a tool to test new ideas, from reinforcement learning agents to LLM characters.

Research CV GitHub
IEEE CoG 2026
Short paper · Madrid
P4 · Tofino
Line-rate telemetry
30+
Plugins audited
2
RL engines
Al Muqshith Shifan portrait
AL · Muqshith オシャワ、オンタリオ 

      

        
        
      

    

  


  
  

    
    

      

        
研究 Research / 01

        
Where the Work Is.

      

      
Code & Sorcery Lab · Ontario Tech

    


    

      
      

        

          
論文 / 001

          
 Presented · IEEE CoG 2026 · Madrid

        

        

          
RIDGE: State-Conditioned Reward Blending for Behavioral Coverage in Deep RL Game Agents

          
AL Muqshith Mohammed Shifan · Kevin Christopher Chua · Ali Neshati · Loutfouz Zaman · Cristiano Politowski

          

            One PPO agent blends four persona rewards: Explorer, Survivor, Craftsman, and Warrior. The game state sets the blend. The agent changes its play style in one episode. This removes the need to train one agent for each persona. A multi-head critic keeps each value function stable while the blend changes. RIDGE is the only condition that reaches both wood-tier crafting branches on Crafter.
          

          

             README
             Camera-ready
             Code
          

        

        

          Collage of Al Shifan on stage at IEEE CoG 2026 with slides on cross-branch coverage, the blend method, and related work
          Al Shifan presents the RIDGE method slide: four activations in, four normalised weights out
          Al Shifan presents the RIDGE results slide with the cross-branch coverage numbers
          Al Shifan at the podium on the last slide, with a QR code to the GitHub repository
          
IEEE CoG 2026 · Madrid · RIDGE short paper

        

      


      
      

        

          
論文 / 002

          
 Draft · Systematic Mapping Study

        

        

          
Emotional Memory in LLM-Driven Non-Player Characters: A Systematic Mapping Study

          
In progress · 66 records screened · 30 primary studies

          

            The standard memory stream for LLM agents ranks past events by recency, importance, and relevance. It does not use emotion. Psychology shows that emotion controls what we keep and what we recall. This study asks two questions. Do LLM NPCs link a stored emotion to a memory? Is that link based on a psychological method? Only 3 of 30 studies do both.
          

          

             Request Preprint
          

        

      

    


    
    

      
01
Persona Stability Inside the Model
Keep authored traits stable at the activation level, not in the prompt. Small models that can run inside a game.

      
02
Emotion-Grounded NPC Memory
A believable NPC must recall the past across a full game. Emotion is the index that decides when a memory is available. Emotion is also a value that the game can write.

      
03
Playtesting Agents With Range
One agent, many play styles. One policy gives behavioural coverage. There is no need for one agent for each persona.

    


    
    

      
Supervised by
        Dr. Cristiano Politowski · primary · Empirical Games / Code & Sorcery Lab
      

      

        Dr. Loutfouz Zaman · Game Development & Games User Research · Dr. Ali Neshati · Human‑Computer Interaction / UXID Lab
      

    

  


  
  

    
    

      

        
私について About / 02

        
A Short Introduction.

      

      
B.IT › MSc · Ontario Tech

    


    

      

        
Who I Am

        

          I am a Master’s student in Computer Science at Ontario Tech University. My work spans networking, security, software design, game development, and AI. I build reinforcement learning agents that play games. I build LLM characters that remember and stay in character. I also build and secure the networks that research runs on.
        

        

          My bachelor’s degree is in Networking and IT Security. I moved to Computer Science to branch out. I want to learn the current work in AI and LLMs in depth. This improves my skills in software design and in AI.
        

        

          I am not a game developer. I am a software designer who uses games as a tool. Games are a harness to test and develop new ideas and concepts.
        

        

          With Dr. Cristiano Politowski I build research tools: PEAK, RIDGE, TAST, and EMBR. With Dr. Loutfouz Zaman and Dr. Ali Neshati I study the human side: how agents show behaviour and how people interact with them. My background in networking and security is why I prefer reproducible systems and adversarial benchmarks.
        

        

          IEEE CoG 2026 first author · Dean’s List 2026 · TA: Web Development, Advanced Networking I, Networking I & II · P4 Programmable Networks RA
        

      


      
    


    

    

      
/ Technical
Palette

      

        
AI & RL

        
    
          
  • PyTorch, Gymnasium
    • 
          
  • Stable-Baselines3 · PPO
    • 
          
  • LLMs · Ollama, Gemini
    • 
          
  • Activation steering
    • 
          
  • Hydra experiment mgmt
    • 
        

      

      

        
Games

        
    
          
  • Crafter env
    • 
          
  • Custom 2D engines
    • 
          
  • Procedural gen (graphs)
    • 
          
  • Persona architectures
    • 
          
  • Playtesting agents
    • 
        

      

      

        
Software

        
    
          
  • Python · C / C++
    • 
          
  • Flask · REST · Three.js
    • 
          
  • React · Node.js
    • 
          
  • SQL / MySQL
    • 
          
  • Docker · GitHub Actions
    • 
        

      

      

        
Systems & Sec

        
    
          
  • Intel Tofino P4
    • 
          
  • Cisco IOS · CCNA
    • 
          
  • Linux / Unix admin
    • 
          
  • Wireshark · Nmap
    • 
          
  • Adversarial benchmarks
    • 
        

      

    


  


  
  

    
    

      

        
通信 Networks & Security / 03

        
Wires & Firewalls.

      

      
B.IT Networking & IT Security · CCNA

    


    

      

        
01 · Data planes

        
P4 on Intel Tofino

        
Built research infrastructure on Intel Tofino programmable switches. Wrote P4 data-plane programs that extract per-flow telemetry at line rate.

        
P4 · Tofino · Telemetry

      

      

        
02 · Security

        
Vulnerability Assessment

        
Assessed a WooCommerce and Stripe stack for YYC Beeswax. Audited more than 30 plugins. Wrote the hardening checklist the team uses as its pre-deploy gate.

        
WooCommerce · Stripe · Hardening

      

      

        
03 · Operations

        
Lab Network Operations

        
Ran a teaching lab network for more than 50 daily users. Cut downtime by about 20%. Maintained the DHCP, DNS, and NTP services and more than 20 workstations.

        
DHCP · DNS · NTP · Linux

      

      

        
04 · Teaching

        
Networking I, II, and Advanced I

        
Teaching assistant for three networking courses. Labs on TCP/IP, subnetting, routing protocols, and enterprise firewalls for about 120 undergraduates a term.

        
TCP/IP · Routing · Firewalls

      

    

  


  
  

    
    

      

        
作品 Projects / 04

        
Research & Code.

      

       Code & Sorcery Lab
    


    


      

        

          
/ 001 · Engine

          
PEAK · Platform Engine

          

            A deterministic, high-performance DRL benchmark engine for 2D platformers. It has SMB1-style physics, dual spatial hashing (about 1000 steps per second), and a modular persona reward system. Design a level and train an agent in the same tool.
          

          

            PythonGymnasiumPPOHydra
          

          

            RESEARCH · ACTIVE
            

              
              
            

          

        

        

          PEAK logo: a pixel-art mountain of controller buttons above the words PEAK 2D Platformer Engine
          PEAK Mega Man-style stage with the debug overlay: hit boxes, ray casts, and a boss health bar
          
        

      

      

        

          
/ 002 · Companion AI · Sneak peek

          
FATHOM · Helpful, Not Omniscient

          

            An AI companion that lives in the Subnautica PDA and is built to stay quiet. It watches oxygen, depth, the dark, and what circles the player. It speaks only when a line changes what the player does, in the original PDA voice. The research question: can an adaptive AI guide a lost diver and keep the fear of survival horror?
          

          

            Unreal 5.6UE4SS LuaOllamaPersona steering
          

          

            CSER 2026 FALL · SNEAK PEEK
            

              
              
            

          

        

        
FATHOM dashboard: vitals, situation vector, persona, range rings, oxygen and depth charts, and PDA transmissions

      


      

        
        
/ 003 · Paper

        
RIDGE · Reward Blending

        
One PPO agent blends four persona rewards from the game state with smooth sigmoids. It is the only condition that reaches both wood-tier branches on Crafter. Presented at IEEE CoG 2026.

        
PyTorchCrafterMulti-head critic

        

          IEEE COG 2026
          

            
            
          

        

      


      

        
/ 004 · Tool

        
TAST · Trait Activation Steering

        
Keep small language models in character across long adversarial conversations, without fine-tuning. It uses cartridges, capping, linear axes, and a rotational dial. Three local LLMs judge the results.

        
PyTorchActivation steeringLLMs

        

          RESEARCH · LOCAL · REPO SOON
        

      


      

        
/ 005 · Middleware

        
EMBR

        
Emotional Memory for Believable Roleplay. NPCs remember what you did and how they feel about it. They give a different answer to a gift and to a betrayal. Emotion is the recall index.

        
PythonAffectAgentic memory

        

          ACTIVE
          

            
            
          

        

      


      

        
/ 006 · Hackathon

        
Gestura

        
Real-time AI body mapping for telehealth. Computer vision and a 3D digital twin turn gestures into structured clinical data. HackHive 2026 · Code & Sorcery Students.

        
FlaskCVThree.js

        

          HACKHIVE 2026
          

            
            
          

        

      


      

        
/ 007 · Live

        

          
Latest GitHub Push

          
Fetching @Al-Scripting…

        

      


      

        

          
/ 008 · Sneak peek

          
Something New Is in the Lab

          
A new project is in development at Code & Sorcery Lab. It is not public yet. Ask me about it.

        

        

          IN THE LAB · 2026
          Ask Me
        

      


    

  


  
  

    
    

      

        
コードと魔術 Code & Sorcery / 05

        
The Lab.

      

       Code-SorceryLab
    

    
Game development, software engineering, and AI meet here. These are the 6 tracks I work in, from cpoli.live/projects.


    

      
        
LLM · Landscape

        
The LLM-NPC Landscape

        
A systematic map of how large language models drive non-player characters, and where the gaps are.

      
      
        
LLM · Memory

        
Emotional Memory in Generative Characters

        
Do LLM characters remember what was important to them? Is that memory based on psychology?

      
      
        
LLM · Stability

        
Staying in Character

        
Keep a character consistent over a long conversation. Work inside the model, not in the prompt.

      
      
        
RL · Coverage

        
One Agent, Many Play Styles

        
Get many play styles from one playtesting agent. Do not train one agent for each style.

      
      
        
RL · Engine

        
PEAK: Authoring and Training in One Loop

        
An open-source 2D platformer engine. Design a level and train an agent on it in the same tool.

      
      
        
Lab · Hackathon

        
Code & Sorcery Students at HackHive 2026

        
Team CSS built Gestura. It identifies body parts and helps a medic and a patient communicate, also across languages.

      
    

  


  
  

    
    

      

        
道 Journey / 06

        
The Path So Far.

      

      
2016 › Present

    


    

      

        
SEP - DEC 2026

        

        

          
Teaching Assistant · Web Development and Advanced Networking I

          
Ontario Tech University

          
Web Development with Randy and Dr. Cristiano Politowski. Advanced Networking I with Brent and Josh.

        

      

      

        
SINCE SEP 2025

        

        

          
Graduate Research Assistant

          
Ontario Tech University · Code & Sorcery Lab

          
Research on LLM game characters, reinforcement learning playtesting agents, and the tools behind them. Supervisors: Dr. Cristiano Politowski (primary), Dr. Loutfouz Zaman (Games User Research), and Dr. Ali Neshati (Human-Computer Interaction, UXID Lab).

        

      

      

        
SEP 2025 - MAY 2028

        

        

          
MSc Student, Computer Science (Software Design)

          
Ontario Tech University · expected graduation May 2028

          
First author on RIDGE (IEEE CoG 2026). I build TAST and EMBR. I do a systematic mapping study on emotional memory in LLM NPCs.

        

      

      

        
2024 - 2026

        

        

          
Teaching Assistant · Networking I & II

          
Ontario Tech University

          
Ran labs on TCP/IP, subnetting, routing protocols, and enterprise firewalls for about 120 undergraduates.

        

      

      

        
JUN - SEP 2025

        

        

          
Network Research Assistant

          
Ontario Tech University

          
Built research infrastructure on Intel Tofino P4 programmable switches. Wrote data-plane programs that extract per-flow telemetry at line rate.

        

      

      

        
JUN - SEP 2025

        

        

          
Cyber Security Analyst

          
YYC Beeswax Ltd · Remote

          
Did vulnerability assessments on a WooCommerce and Stripe stack. Audited more than 30 plugins. Wrote a hardening checklist that the engineering team uses as its pre-deploy gate.

        

      

      

        
2025

        

        

          
Bachelor of IT, Networking and IT Security

          
Ontario Tech University

          
Networking, systems, and security. This is the base for my move to Computer Science and AI.

        

      

      

        
SEP 2024 - JUN 2025

        

        

          
Networking Lab Technician

          
Ontario Tech University

          
Cut lab network downtime by about 20% for more than 50 daily users. Maintained more than 20 workstations and the DHCP, DNS, and NTP services.

        

      

      

        
2022 - 2023

        

        

          
Programming Instructor

          
Kids Great Minds · Whitby

          
Taught digital technology in hands-on courses. Mentored students on personal projects.

        

      

      

        
2018

        

        

          
IT Field Service Staff

          
TDSB · Scarborough

          
Kept the asset inventory for more than 2,000 laptops. Ran factory-reset and QA workflows that follow the board compliance standards.

        

      

    

  


  
  

    
    

      

        
便り Notes / 07

        
Latest Updates.

      

       Follow on LinkedIn
    

    

  


  
  

    
    

      

        
連絡 Contact / 08

        
Work With Me.

      

      
Reply within 24 hours

    


    

      

        
Get in Touch

        
I look for research collaborations and engineering roles in software design, network engineering, security, game AI, and LLM systems.

        
      

      

        

          
Reach Out

          
Email for proposals, LinkedIn for chats, GitHub for code.

        

        

           Send an Email
          

             LinkedIn
             GitHub
             CV
          

        

      

    

  
















  ADMIN MODE
  
  





  




  




  




  

    
    

      読み物
      README
    

    
Fetching README…