0
Initializing
45.5274°N — 12.6014°E  ·  Veneto, Italy
Dario Leonardo Arcan  /  AI & Data Science  /  World-Ranked Competitor

I optimize everything.
Starting with myself.

World-ranked across 5+ competitive titles. Now I channel that same obsessive pattern recognition into AI systems, reinforcement learning, and data science.

Scroll
01Identity
The same mind that reached
world rankings in competitive
gaming now builds systems
that think.

Most people treat gaming as entertainment. I treated it as a laboratory for decision-making under uncertainty — reading opponents, optimizing strategies in real time, finding edges invisible to everyone else. That's what took me to the Hearthstone Master Tour, to European and global leaderboards across multiple titles.

That cognitive toolkit doesn't disappear when you switch domains. It transfers directly into reinforcement learning, probabilistic modeling, and system design. I'm now a BSc AI & Data Science student at H-FARM College, and every project I ship carries the same intensity: find the optimal solution, execute precisely, iterate relentlessly.

5+
Games, World-Ranked
7+
Shipped Projects
RL
Core Focus
24
Years Old
02Competitive Record
#MT
World Championship
Hearthstone
Qualified and competed at Hearthstone Master Tour 2021 — Blizzard's premier global competition. Thousands of players, one stage. The highest level of competitive card game play in the world.
#2
World
Squad Busters
Peaked at #2 globally and #2 in Italy in Supercell's real-time strategy brawler. Rapid decision-making and adaptive strategy under chaotic conditions.
#3
World
Mo.Co
Peaked at #3 in the world. Pure mechanical precision combined with strategic depth — consistent top-tier performance from day one.
#16
Europe  ·  #3 Italy
Teamfight Tactics
Top 16 in Europe, #3 in Italy in Riot Games' auto-battler. Economy management, probabilistic item optimization, and real-time composition pivoting — the closest any game gets to applied statistics.
<1%
Global  ·  Multiple Titles
Cross-Game Dominance
Consistently ranked in the top <1% globally across every competitive title I've touched. The skill that transfers isn't game-specific — it's the ability to decode systems, find hidden structure, and exploit it faster than anyone else.
◆
Why this matters for AI: Competitive gaming at this level is applied decision theory — opponent modeling, expected value calculation, real-time Bayesian updating, and strategy under incomplete information. These are the same primitives that power reinforcement learning and intelligent systems.
03Selected Work
01
Adaptive AI
Coco
An AI companion that doesn't just respond — it adapts emotionally. Coco runs a Finite State Machine with 6 mood states (happy, sad, curious, anxious, calm, excited) that shift based on conversation context, time patterns, and user behavior history.
Architecture: Three-tier memory system (short-term buffer → session context → long-term persistence). Hybrid LLM routing with automatic fallback chain: Claude API → OpenAI → local mock. Built on Strategy, Observer, State, and Repository design patterns. 38/38 unit tests passing. Python 3.12, pyproject.toml managed.
PythonFSMClaude APIOpenAIDesign PatternsUnit Testing
→
02
Reinforcement Learning
RL Mario Agent
A PPO agent trained entirely from raw pixel input to beat Super Mario Bros World 1-1. No hand-crafted features, no human demonstrations — just a neural network learning to play through trial and error on 240×256 RGB frames.
Pipeline: Frame preprocessing (grayscale, 84×84 resize, 4-frame stack) → CNN feature extractor → PPO policy with entropy bonus. Trained on RTX 5070 (Blackwell sm_120 architecture, required PyTorch nightly cu128). Diagnosed and resolved policy collapse through learning rate scheduling and entropy coefficient tuning. ~60% World 1-1 completion rate. Full PDF report built with ReportLab.
PyTorchStable-Baselines3PPOComputer VisionCUDAReportLab
→
03
Swarm Robotics
NECTAR
A precision agriculture platform built around a mother drone + micro-drone swarm architecture. Each micro-drone handles a specialized task — seeding, pollination monitoring, or targeted pesticide application — coordinated by a central command unit.
Concept: Bio-inspired swarm coordination with task allocation algorithms, GPS-guided waypoint navigation, and real-time sensor feedback loops. Developed for a robotics hackathon (3-minute pitch format, 30 groups). Delivered a complete pitch package: business plan, market analysis, competitive landscape, and technical architecture deck.
Swarm IntelligenceRoboticsPrecision AgPitch StrategyBusiness Planning
→
04
Game Theory
CINQUE
A zero-luck 1v1 strategy game inspired by the Colonel Blotto problem. Two players simultaneously allocate a 100-point budget across 5 lanes with asymmetric values (1 through 5). Whoever commits more to a lane wins its value. Pure cognition — no dice, no draws, no RNG.
Stack: React Native + Expo frontend, Firebase Realtime Database for synchronous multiplayer state. Commit-reveal anti-cheat system — both players submit encrypted allocations before either is revealed. Node version conflicts resolved via nvm-windows; clean project architecture at cinque-fresh.
React NativeExpoFirebase RTDBGame TheoryCryptographic Commit-Reveal
→
05
Game Development
FOSSALTA77
A cyberpunk-themed multiplayer card game — designed, built, and deployed as a single HTML file. 20 unique cards with distinct mechanics, cyberpunk visual identity, and real-time multiplayer through Firebase.
Shipped: Firebase Realtime Database for game state sync, Firebase Hosting for deployment (free Spark plan). Entire game logic, UI, and styling in one self-contained file. Card balance designed through iterative playtesting. Live and playable.
Vanilla JSFirebase RTDBFirebase HostingGame DesignCard Balance
→
06
Optimization
LDM-Optimizer
A 3D bin-packing algorithm that minimizes Loading Meters in semi-trailer trucks. Took over a legacy system and rewrote the core packing logic from scratch in TypeScript.
Algorithm: Row-First placement discipline combined with Guillotine residual fill and Simulated Annealing metaheuristic. Delivered as a drop-in REST endpoint replacement with full ARCHITECTURE.md documentation. Handles real-world constraints: weight distribution, stackability limits, mixed box dimensions.
TypeScriptBin PackingSimulated AnnealingREST APIOptimization
→
07
Corporate Workshop · 1st Place
FiberCop — Issue Priority Mapper
Won 1st place at the FiberCop corporate volunteering workshop at H-FARM College (Dec 2025). Built a software tool that maps community infrastructure problems, visualizes them geographically, and assigns priority scores for intervention scheduling.
Concept: Problem intake system with severity classification, geographic clustering, and weighted priority scoring algorithm. Designed to help FiberCop field teams allocate resources to the highest-impact issues first. Presented and defended in a competitive team pitch format.
Problem MappingPriority Scoring1st PlaceTeam PitchH-FARM
→
04The Path
2021
Competitive Peak
Hearthstone Master Tour qualifier. Top rankings across TFT, Squad Busters, Mo.Co, and more. Built a Twitch presence as mrdarione. Learned what it means to perform when it actually matters.
2023
The Pivot
Luxury retail at Karl Lagerfeld and Trussardi. Chemical Engineering at University of Padova. Searching for where strategy, technology, and building converge — then finding it in AI.
NOW
Building & Winning
H-FARM College, BSc AI & Data Science. Won 1st place at the FiberCop corporate workshop (Dec 2025). Shipping projects obsessively. Targeting elite international AI/DS roles.
05Arsenal
PythonPyTorchStable-Baselines3 ReactReact NativeTypeScript FirebaseClaude APIOpenAI API KNIMEGitLinux Probability & StatisticsGame Theory Reinforcement LearningCUDA PythonPyTorchStable-Baselines3 ReactReact NativeTypeScript FirebaseClaude APIOpenAI API KNIMEGitLinux Probability & StatisticsGame Theory Reinforcement LearningCUDA
06Résumé
Education
BSc Artificial Intelligence & Data Science
H-FARM College — University of Chichester · Venice, Italy
Sep 2024 — Present
Core modules: Linear Algebra, Probability, Algorithms & Data Structures, Data Science. Active participant in extracurricular workshops and tech challenges.
Diploma — Automation & Robotics
ITIS Vito Volterra · San Donà di Piave (VE), Italy
2015 — 2020
Work Experience
Sales Assistant
Karl Lagerfeld
Nov 2024 — Present
Customer-facing sales in luxury fashion. Building client relationships, managing high-value transactions, consistent performance in a high-standard retail context.
Sales Assistant
Trussardi
May 2024 — Oct 2024
Sales and client engagement in premium fashion. Multi-lingual customer interactions in an international retail environment.
Sales Assistant
Karl Lagerfeld
2023 — 2024
First experience in luxury retail. Developed foundational client management and high-pressure communication skills.
Achievements
1st Place — FiberCop Corporate Workshop
H-FARM College · Venice, Italy
Dec 2025
Won a competitive corporate workshop organised by FiberCop. Developed and presented a software tool for mapping infrastructure issues and assigning intervention priorities.
Languages
ItalianNative
RomanianNative
EnglishB2
SpanishB1
FrenchA1
GermanA1
Certifications
Microsoft Power BIMicrosoft
Introduction to AICisco Networking Academy
Python Essentials 1Cisco Networking Academy
Claude API & Agent DevelopmentAnthropic · In Progress
D.L. ARCAN
07Contact
Let's build
something extraordinary.

Looking for internships, research collaborations, and ambitious projects in AI, data science, and intelligent systems.

GitHub LinkedIn Email