Hi, nice to meet you! I'm Carlo Lepelaars and I enjoy exploring Artificial Intelligence (AI), Mathematics, Quantum Computing, Investing and Music. I love reading, thus I've compiled a list of my readings [here](/books).

  • I used to be an aspiring full-time drummer and played in various bands. You can still view live videos [here](https://www.youtube.com/watch?v=HLFZ8z8GSGI&list=PLx5WERJ3t5uXYIJ39ZFj5VfGEdCWp_scG). During this time I was studying Commercial Economics in The Netherlands.
  • I developed a fascination for AI, switched my focus and graduated in Data Science at the Eindhoven University of Technology (TU/e) in The Netherlands.
  • A major part of my education as a data scientist has been through working on data science competitions like [Kaggle](https://www.kaggle.com/carlolepelaars) and [Numerai](https://numer.ai). Creating and contributing to open source software has also been crucial. Through Numerai, I met [Jason Rosenfeld](https://crowdcent.com/about) and started working at [CrowdCent](https://crowdcent.com).
  • I play guitar and love [sim-racing](https://en.wikipedia.org/wiki/Sim_racing)!
Carlo Lepelaars

CrowdCent Logo (2022-now)

I work for [CrowdCent](https://crowdcent.com) as a (founding) data scientist on financial machine learning (ML), large language models (LLMs) and MLOps. CrowdCent is a startup working on investment strategies driven by AI trained on investment community data. CrowdCent is collaborating with investment community [SumZero](https://sumzero.com) on a fund. CrowdCent also runs a fund where we build ML models for the financial data science tournament [Numerai](https://numer.ai) and stake on them for investors.

Past Work


DeepNewz (2024)

Consultancy on:

  • Large Language Models (LLMs) workflows
    • Prompt Engineering
    • Automated News Story Writing
    • Topic Classification
    • Language Translation
  • Prediction Markets
    • Prediction Market Generation
    • Prediction Market Quality Scoring

VBTI (2020-2021)

Computer Vision for agriculture:

  • Rotational Estimation
  • Object Detection
  • Segmentation
  • Keypoint Detection

Weights & Biases (2020-2021)


Open Source Projects


numerblox

scikit-learn components for [Numerai](https://numer.ai/). We use NumerBlox within CrowdCent as part of our own Numerai infrastructure.

blackscholes

Black-Scholes (financial option pricing) calculator. A lightweight package to determine option prices with [extensive documentation](https://carlolepelaars.github.io/blackscholes/3.the_greeks_blackscholes) on the math behind the Black-Scholes model.

fh-plotly

[Plotly](https://plotly.com) to [FastHTML](https://www.fastht.ml) conversion plugin.

ghostnet_tf2

At the time one of the first implementations of the [GhostNet](https://arxiv.org/abs/1911.11907) CNN architecture in Tensorflow 2.0.

Open Source Contributions

scikit-lego

Extra blocks for scikit-learn pipelines. Contributions included generalizing components to multi-output, custom arguments and ensuring scikit-learn compatibility.

FastHTML-Gallery

A gallery of [FastHTML](https://fastht.ml) examples. I contributed [Plotly visualization examples](https://gallery.fastht.ml/vizualizations/bloch_sphere/display), an [audio player](https://gallery.fastht.ml/widgets/audio/display) and [PDF viewer](https://gallery.fastht.ml/widgets/pdf/display).

narwhals

A compatability layer between DataFrame libraries (Polars, Pandas, PyArrow, Dask, etc.). I contributed the `skew` function and documentation.

umap-learn

UMAP Dimensionality Reduction. I helped out with some minor contributions for ensuring scikit-learn compatibility.

embetter

Embedding models easily accessible as scikit-learn components. Contributed minor bug fixes.

numerai-tools

Tools and utilities for Numerai. I helped out with metrics and dependency management. [NumerBlox](https://github.com/crowdcent/numerblox) uses numerai-tools to ensure consistency with how Numerai calculates payouts.