About Waeolu

Where machine learning meets investment practice

Waeolu started in 2020 with a straightforward premise: the gap between academic ML research and what investment professionals actually use was embarrassingly wide. We build the workshops that close it.

Waeolu workshop participants working through ML investment models

Our story

Built from a real frustration

Quantitative finance courses taught theory on clean datasets. Actual portfolio teams worked with messy, delayed, and expensive data. Nobody seemed bothered by the distance between those two realities — so we decided to be bothered by it.

Each workshop at Waeolu runs on live data pipelines, realistic transaction cost models, and scenarios where the algorithm behaves unexpectedly at exactly the wrong moment. Participants leave knowing what breaks and why.

Cambridge, Ontario turned out to be a decent place to build this. The city has a serious manufacturing and engineering culture — people here are used to the idea that systems need to be tested before they're trusted.

14+ Workshop modules covering signal generation, backtesting, risk management, and execution logic

6 Instructors with backgrounds split between academic ML research and institutional portfolio management

4 Cities with active regional cohorts, each adapted to local market data and regulatory context

The people

Practitioners who still read papers

Every instructor on the platform has worked with real capital — not just simulated portfolios. The academic side matters too: the field moves fast enough that ignoring research means teaching outdated methods within eighteen months.

Small teams mean participants get direct access. Questions don't go to a forum queue.

Waeolu instructors in a collaborative session

Cambridge, ON

Lead Instructor

Tobias Wrenfield

Spent eight years at a Toronto-based quant fund before moving to education. Specializes in factor model construction and the practical limits of momentum strategies.

ML Architecture

Priya Nambiar

PhD in statistical learning from Waterloo. Focuses on the gap between offline model performance and what actually happens when a model touches live order flow.

Risk & Execution

Alistair Brogue

Former risk analyst at a mid-size asset manager. Builds the modules on position sizing, drawdown attribution, and why most backtests overfit to the last market regime.

Data Engineering

Saoirse Denvir

Handles the infrastructure side — real-time feeds, feature pipelines, and the unglamorous work of making sure data arrives correctly before any model touches it.

How we work

Structure that matches the actual job

Financial ML breaks in predictable ways — data leakage in the feature pipeline, transaction costs that turn a profitable signal into a losing strategy, regime shifts that invalidate three months of tuning. Workshops here are structured around those failure modes, not around success stories.

Each module moves from concept to implementation to stress test. Participants run the code, break it deliberately, and learn to read what the failure is telling them. That sequence is slower than a lecture, and considerably more useful.

Concept grounding

Each topic opens with the specific problem it solves — not with history or motivation. Context comes after the mechanism is clear.

Guided implementation

Participants build the component themselves with structured scaffolding. Reading finished code is not the same as writing it under constraint.

Deliberate stress testing

Models run against adversarial data — thin liquidity, missing values, sudden volatility spikes. Robustness is tested, not assumed.

Group critique

Approaches are compared across cohort members. Different design decisions produce different failure modes — seeing several at once accelerates calibration.

What we hold to

Principles that survived contact with reality

These aren't aspirations. They're constraints we've found ourselves returning to when a workshop design goes wrong.

Workshop environment showing collaborative ML analysis work

Failure is the curriculum

Workshops are designed around what goes wrong, not what goes right. A model that only succeeds on clean data teaches very little about real deployment conditions.

Specificity over scope

A workshop that covers everything covers nothing usefully. Each module has a narrow focus and goes deep enough to be genuinely applicable.

Small cohorts, direct feedback

Groups stay small enough that instructors can review individual implementations. Generic feedback on generic work is not worth anyone's time.

Honest about what takes time

ML competence in investment contexts develops over months of iteration, not days of instruction. Workshops accelerate the learning — they don't replace the time it takes to build judgment.

Regional context matters

Canadian equity and fixed income markets have specific characteristics — settlement conventions, liquidity profiles, regulatory constraints — that generic US-focused curricula quietly ignore.