Data Lab
12
models & tools
From pricing models and ETL pipelines to small trading experiments and research notebooks.
Yohan Zytoon
Quantitative Finance · Data & ML
Montréal, Québec, Canada
I like building tools that make data easier to understand and work with, everything from prediction models and research notebooks to small interfaces that make complex ideas feel intuitive.
At Desjardins Insurance, I work on pricing ML models and end-to-end data pipelines used in real decisions, which taught me the value of clean data, reproducibility and thoughtful engineering.
I enjoy working where data, software, and ML meet, whether that’s experimenting with trading ideas, exploring new models, or building tools that help make decisions clearer.
Data Scientist @ Desjardins
Prev BI @ BRP
Decision Stack
Bringing together ML, data engineering and intuitive interfaces to support better decisions.
Prediction models
Data pipelines
ML systems
Interactive tools
Currently building
Introduction
About
A snapshot of how I think about quantitative finance, data and building tools that actually get used.
I’m an undergraduate in Mathematics & Computer Science at the Université de Montréal, interested in machine learning and data-driven systems. I enjoy taking a vague idea (“we should try this”) and turning it into a real project… usually powered by an unhealthy amount of caffeine and one or two existential crises.
Outside class, I’m the Treasurer of UdeM AI and I practice Brazilian Jiu-Jitsu (blue belt) — which turns out to be a surprisingly good metaphor for decision-making under pressure. I also played rugby with the Carabins (UdeM), and I enjoy thinking games like chess and poker, mostly to learn how humans (including me) make questionable decisions with confidence.
Outside the screen, I split my time between BJJ, gym, chess, poker and discovering Montréal cafés and restaurants with my girlfriend. I like learning things just for the sake of it, whether it’s a concept, a model, or a hobby I’ll pretend I have time for. It keeps life interesting.
Profile at a glance
- Building things I probably shouldn’t enjoy this much.
- Calling differential equations “fun” unironically.
- Resisting the urge to throw my laptop when trading or debugging code.
- Letting ChatGPT fix the mess when I break everything.
Areas of interest
Quantitative & algorithmic trading
Quant & risk research
Data Science & Machine Learning
AI research
Current Focus
Now
What I’m actively working on and thinking about right now.
- Building ML pricing models and ETL pipelines at Desjardins Insurance and trying not to break anything on production days.
- Running small trading experiments, market regimes, volatility signals, execution tweaks, all the fun stuff.
- Designing personal research tools that make experimenting easier: clean data loaders, dashboards and sandbox notebooks.
- Building a clearer decision-making framework so my trading choices rely more on data and less on vibes.
- Studying for CFA Level I (May 2026) and discovering new ways derivatives make my brain hurt.
- Reading game theory, poker and trading psychology so my decision-making becomes slightly less chaotic.
Academic Background
Education
Formal training that underpins my quantitative and computational work.
Université de Montréal
B.Sc. in Mathematics & Computer Science
- Digging deep into stochastic processes, numerical optimization and the math that powers machine learning models.
Collège Ahuntsic
Diploma of College Studies (DEC) — Accounting and Management
- Learned to love balance sheets, budgeting and the business side of every engineering decision.
Professional Journey
Experience
Roles where I applied mathematical reasoning, analytics and financial intuition.
Data Scientist — Advanced Analytics Intern
Desjardins Assurances générales (DAG)
- Built pricing sandboxes where new ML features take a beating before actuaries ever see them.
- Instrumented pipelines with the kind of telemetry that lets us replay any run without tears.
- Translated experiments into capital decisions for treasury and underwriting leads.
Business Intelligence Intern
Bombardier Recreational Products (BRP)
- Designed dashboards that tracked margin swings and gave leadership a “no drama” view of supply.
- Automated the SQL + Power BI handoffs so analysts could stop babysitting refresh buttons.
- Modeled supplier scenarios that fed directly into contract negotiations.
Financial Services Associate
CIBC
- Analyzed client portfolios totaling $800M AUM, focusing on risk exposure and asset mix.
- Improved onboarding flows, reducing average handling time by ~20%.
Customer Advisor
Royal Bank of Canada (RBC)
- Processed high transaction volumes while maintaining strong client satisfaction.
- Identified customer needs and recommended products, contributing to branch growth.
Independent Work
Projects
Self-initiated work that reflects how I explore ideas and build tools beyond the classroom.
Multi-Agent Liquidity Lab
PyTorch · Ray RLlib · PyG · Game Theory
- Frames quoting across venues as a repeated stochastic game where agents balance spread capture, inventory drag and signalling risk.
- Shares state through a graph neural network so agents digest correlated limit-order books without leaking raw order flow.
- Trains cooperative and adversarial policies that learn when to take spread, yield, or shadow hedge across venues.
Athena HopeBridge — Donor Experience Platform
Morgan Stanley CodeToGive · React · FastAPI · Django · RAG
- Designed a multilingual, trauma-aware donor experience with real-time impact stats and “Impact Pathways”.
- Built the full Athena Guide chatbot microservice (FastAPI, embeddings, intent detection, RAG) with crisis overrides.
- Integrated donor recommender, 3-step donation flow, dashboards and future LLM integration roadmap.
Limit Order Book Simulator & Backtester
C++ · STL · CMake · pybind11
- Maintains a price–time priority book with market, limit, IOC and peg orders.
- Streams historical event data to replay venue dynamics at sub-millisecond fidelity.
- Benchmarks market making and stat-arb tactics via Sharpe, drawdown and inventory risk metrics.
Black–Scholes Option Pricing Tool
Python · Streamlit · NumPy · SciPy · Plotly
- Developed a UI where users manipulate inputs and see prices & Greeks update in real time.
- Added Monte Carlo and finite-difference implementations to validate Black–Scholes outputs.
Blockchain Fraud Detection
PyTorch · PyG · Graph Neural Networks
- Implemented GNN models with residual connections to classify transactions as legitimate or illicit.
- Engineered node features and temporal splits to better reflect on-chain dynamics.
- Achieved high precision on fraud detection while controlling for class imbalance.
Additional Technical Projects
Computer vision · NLP · Compilers · Game AI
- Face Detection & Recognition System — real-time OpenCV toolkit for detection and LBPH recognition. GitHub
- VaR–ML–DL Comparison — evaluating classical, machine learning and deep learning VaR models on the same portfolio. GitHub
- Career Compass — Resume Matcher — embeddings-based resume–job alignment. GitHub
- Email Spam Classifier — classic NLP pipeline reaching ~97% accuracy on spam vs. ham. GitHub
- Lisp-like Interpreter & Minimax Tic Tac Toe — experiments in interpreters and game search in Haskell & C++. Interpreter · Tic Tac Toe
Deep Dives
Case Study
A look at an exploratory research sandbox I’m building to study liquidity, competition and execution using multi-agent reinforcement learning.
Multi-Agent Liquidity Game (Research Sandbox)
Problem
Liquidity isn’t set by a single model—it’s the emergent result of makers, takers and arbs reacting to each other and to changing regimes. I wanted a sandbox to study those dynamics with controllable synthetic order books before touching any production setting.
Approach
The sandbox combines:
- Synthetic multi-asset/venue limit order books with normal vs. stress regimes and configurable noise flow.
- Role-specific agents (maker, taker, arbitrageur) trained with a minimalist decentralized PPO loop.
- Graph-based observations (GraphSAGE encoder) so agents see spreads/depth/vol/imbalance as structured node/edge features instead of raw events.
- Reward shaping that balances PnL with inventory risk, execution quality and unhedged exposure to test how behaviour shifts across regimes.
Current status
The environment, PPO trainer, and scripts for single-/multi-agent runs are in place. Recent runs show sensible spread widening in stress regimes, but the arbitrage agent still loses money and taker completion incentives need tuning.
Next steps
Add richer regime calibration from the bundled synthetic datasets, tighten taker completion rewards, rebalance arb penalties/close bonuses, and experiment with a centralized critic for stability. Longer term: integrate attention-based GNN layers and log more reward components for diagnosis.
Interactive
Demos
Small, self-contained tools that show how I think with code. Everything below runs client-side.
Black–Scholes Pricing Mini Demo
Quick computation of theoretical option prices and key sensitivities using Black–Scholes.
Adjust the inputs above to compute a price and Greeks.
Why I like this demo:
It’s simple enough to live in a browser, but rich enough to talk about:
It’s simple enough to live in a browser, but rich enough to talk about:
- How assumptions (lognormal returns, constant σ) limit models.
- How Greeks describe risk in a way that’s teachable to non-quants.
- How to validate pricing engines with alternative numerical methods.
Reading List
Research Articles I Like
A small shelf of DL / RL / game-theory papers I revisit often, plus why they matter to my work.
Proximal Policy Optimization Algorithms
The pragmatic RL workhorse. Clipped objectives and GAE make PPO stable enough for messy environments; it’s my default baseline before trying fancier multi-agent tricks.
Attention Is All You Need
The cleanest blueprint for sequence modelling without recurrence. I borrow its attention patterns when experimenting with order-flow encoding and when comparing graph encoders to sequence models for market microstructure.
Graph Attention Networks
A lightweight way to let graphs learn where to focus via attention over neighbors. It’s my go-to reference before adding relational inductive bias to financial graphs or swapping GraphSAGE for something more expressive.
Toolkit
Skills
What I’m comfortable using day-to-day, grouped by how I actually think about my toolbox.
Programming & Libraries
- C/C++, Python, R, SQL, Java, Haskell
- NumPy, Pandas, scikit-learn
- PyTorch, TensorFlow, PyG
- Git & GitHub workflows
Data & Platforms
- Power BI (DAX), Excel / VBA
- Snowflake, BigQuery, PostgreSQL, MySQL
- Azure ML & analytics tooling
- Dockerized environments
Math & Methods
- Probability & stochastic processes
- Regression & statistical modelling
- Optimization & Monte Carlo simulation
- RNNs & GNNs & RL
Languages
- English — Fluent
- French — Fluent
- Arabic — Native
Foundation
Selected Coursework
Courses that shaped my quantitative and technical foundation.
Machine Learning
Adversarial ML (Audit)
Statistical Learning
Bayesian Statistics
Advanced Probability & Statistics
Stochastic Processes
Real Analysis
Numerical Methods
Algorithm Design
Data Structures
Optimization & Operations Research
Data Science
Data Mining
Database Systems
Derivatives Pricing & Risk Management
Corporate Finance & Accounting
Financial Statement Analysis
Azure AI Fundamentals (Self-study)
Credentials
Certifications
Formal programs that complement my academic background.
CFA Program — Level I Candidate
Exam: May 2026
For Hiring Teams
For Teams & Recruiters
How I can contribute to a trading, product, or research-driven quant team.
- Research & prototyping: I can explore ideas (microstructure, risk, ML) and turn them into reproducible experiments with clear metrics.
- Tooling & automation: I like building small internal tools (dashboards, scripts, simulators) that reduce friction for others.
- Bridging quant & engineering: I’m comfortable thinking in math, code and product at the same time.
- Product-aware thinking: Whether it’s a trading desk, research pod, or data/ML team, I care about how the tools we build plug into real workflows and decisions.
You can get a concise snapshot of my profile in the 1-page CV and more depth (context, projects, outcomes) in the detailed version: