intentional ml
// selected work

Selected work

Generative AI and classic machine learning, shipped for startups and established companies in Canada, the US and Europe. Each project below is labeled as client work or an in-house role.

client projectwork for a client in-house rolework as an employee
case_01 client project generative ai

An AI assistant inside a website builder

Forge and Smith, Vancouver · 2024
problem

Users of Refoundry, Forge and Smith's low-code website builder, needed to search and get answers from its documentation without leaving the product.

approach

I scoped the requirements directly with the client, then built an AI assistant in Python on FastAPI, running on Google Cloud, with a ChromaDB vector database for retrieval and the OpenAI API for answers.

result

Search and documentation Q&A now work inside Refoundry.

python fastapi google cloud chromadb openai api
case_02 in-house role generative ai

Generative AI for legal and tax professionals

Lefebvre SDU, The Hague · 2026 to present
problem

Legal and tax professionals need answers they can trust, which raises the bar for a generative AI assistant.

approach

As Senior AI Engineer I develop GenIA-L, SDU's generative AI assistant, and lead the move to more robust ML infrastructure with LangGraph and MLflow.

python langgraph mlflow
case_03 in-house role classic ml signal processing

Weapon detection from electromagnetic signals

Xtract One, Vancouver · 2020 to 2023
problem

Detect weapons reliably from electromagnetic field data, on embedded hardware with no room for slow or heavy models.

approach

As Lead Machine Learning Engineer I delivered classification and localization models for embedded systems, built xt-training, an internal PyTorch and scikit-learn framework for faster model iteration, and designed a microservice platform to deploy models across gateway systems.

result

Moving the pipeline to MLflow, BentoML and an SQL-based datastore significantly reduced experimentation and deployment time.

pytorch scikit-learn mlflow bentoml sql embedded systems
case_04 in-house role computer vision

Computer vision for esports coaching

Insights.gg, Vancouver · 2018 to 2020
problem

Coaches wanted statistics from gameplay video, not hours of manual review.

approach

I developed and fine-tuned models with TensorFlow, Keras and scikit-learn to extract gameplay patterns from footage, and led the team that built the production deployment on Google Cloud Functions with an API in Go.

result

The analysis tool was adopted by professional Overwatch League teams and freelance Super Smash Bros. Ultimate coaches.

tensorflow keras scikit-learn cloud functions go
// also

Snappet

in-house role · 2025

Productized knowledge-tracing models for hundreds of thousands of students in the Netherlands, on AWS SageMaker, Athena, Lambda and CDK.

MITRE

in-house role · 2017 to 2018

Received the Officer's Award for machine learning and computer vision work on drone imagery for Project MAVEN.

Have a model that needs to ship, or an idea you're not sure about?

Book a 30-minute call See services