Leon Chlon

hi,

leon here.

I create stuff sometimes.

ML researcher, engineer, and AI educator (Apple, Harvard, Cambridge).

✉️ Say hi
⬇️ Bayesian Flashcards

/about me

I’m a Research/ Machine Learning Engineer at Apple and a former Research Fellow at Harvard and Cambridge.

Technologies I’ve worked with recently:

  • Python
  • React.js
  • Java
  • C#
  • PyTorch
  • TensorFlow
  • Apache Airflow
  • Kafka
  • Hive
  • Scuba
  • PyMC4

Off the clock you’ll find me writing papers on multimodal learning, building Bayesian learning tools or offering free AI career coaching for an audience of 150k across TikTok, LinkedIn, and Instagram.

Leon portrait

/experience

Lead Machine Learning Engineer @ Apple

  • Led redesign of entity recommendation pipeline by deploying self-supervised transformers in PyTorch (attention pooling + MMMoE), achieving 25% latency reduction and significant embedding uplift.
  • Collaborated cross-functionally to ship scalable ML infrastructure under tight data constraints.
  • Led redesign of entity recommendation pipeline by deploying self-supervised transformers in PyTorch (attention pooling + MMMoE), achieving 25% latency reduction and significant embedding uplift.
  • Collaborated cross-functionally to ship scalable ML infrastructure under tight data constraints.

Senior Data Scientist (P4 / GF) @ World Bank Group

  • Directed AI time-series strategy for global economic indicators (300k+ series) by integrating LLM-based changepoint scoring and Gaussian Process binary search methods.
  • Delivered 3x processing speed and 30–50% false-positive reduction in anomaly detection stack.
  • Directed AI time-series strategy for global economic indicators (300k+ series) by integrating LLM-based changepoint scoring and Gaussian Process binary search methods.
  • Delivered 3x processing speed and 30–50% false-positive reduction in anomaly detection stack.

Senior Machine Learning Engineer @ TikTok

  • Optimized attention heads for video ranking using multi-task learners with generative transformer-derived features, improving user engagement across 100M+ users.
  • Ran large-scale A/B experiments to tune user retention (staytime, CTR).
  • Optimized attention heads for video ranking using multi-task learners with generative transformer-derived features, improving user engagement across 100M+ users.
  • Ran large-scale A/B experiments to tune user retention (staytime, CTR).

Principal Machine Learning Engineer @ Tailor Bio

  • Built LLM-enhanced scraping pipelines for biomedical patent data (terabytes) using NLP APIs, boosting knowledge graph completeness.
  • Engineered ChemBERTa-based small molecule models with SMINA docking and homology modeling, reducing drug-target prediction error by over 20%.
  • Built LLM-enhanced scraping pipelines for biomedical patent data (terabytes) using NLP APIs, boosting knowledge graph completeness.
  • Engineered ChemBERTa-based small molecule models with SMINA docking and homology modeling, reducing drug-target prediction error by over 20%.

Senior Machine Learning Engineer @ Uber (Careem)

  • Designed real-time geospatial pipelines with Apache Airflow + Kafka on S3; enabled dynamic pricing at scale.
  • Applied cooperative game theory to pricing algorithms, improving vehicle flow by 20%.
  • Designed real-time geospatial pipelines with Apache Airflow + Kafka on S3; enabled dynamic pricing at scale.
  • Applied cooperative game theory to pricing algorithms, improving vehicle flow by 20%.

Research Data Scientist – Infrastructure @ Meta

  • Deployed RL-based beta bug detection in PyMC3, improving crash discovery rate by 5%.
  • Finetuned FAISS NLP transformer embeddings for hate speech detection (15% recall gain).
  • Built predictive dashboards with Hive, Scuba, and fbprophet to monitor code health across Meta apps.
  • Deployed RL-based beta bug detection in PyMC3, improving crash discovery rate by 5%.
  • Finetuned FAISS NLP transformer embeddings for hate speech detection (15% recall gain).
  • Built predictive dashboards with Hive, Scuba, and fbprophet to monitor code health across Meta apps.

Senior Data Scientist – Risk Practice @ McKinsey & Company

  • Advised banks across Europe and Asia on credit scoring under regulatory frameworks; implemented causal ML models improving accuracy by 40%.
  • Led VAR-based economic scenario modeling and designed risk models using socio-economic data.
  • Advised banks across Europe and Asia on credit scoring under regulatory frameworks; implemented causal ML models improving accuracy by 40%.
  • Led VAR-based economic scenario modeling and designed risk models using socio-economic data.

education

Harvard Medical School & MIT
Computational Neuroscience
Postdoc 2018
Postdoctoral Fellowship. Neural data analysis and ML in neuroscience.
University of Cambridge
Applied Statistics & Bioinformatics
PhD 2017
Doctoral thesis on machine learning for biological data; bioinformatics & statistical genetics.
University of Cambridge
Condensed Matter Physics MPhil 2014
Master's research in condensed matter & quantum physics.
University of Warwick
Mathematics & Physics BSc 2013
Bachelor's with emphasis on mathematical modeling and theoretical physics.
LSHTM
Bayesian Statistics & Epidemiology Fellowship 2024
Research Fellow. Advanced Bayesian methods for medical statistics and epidemiology.

/select papers

Multitaper Infinite Hidden Markov Model for EEG

IEEE Conference Paper

Advanced signal processing techniques for electroencephalography analysis using infinite hidden Markov models with multitaper spectral estimation.

Read Paper

Robust Multimodal Learning via Entropy-Gated Contrastive Fusion

arXiv Preprint

Novel approach to multimodal learning using entropy-based gating mechanisms for improved contrastive fusion across different data modalities.

Read Paper

Master Regulators of Oncogenic KRAS Response in Pancreatic Cancer: An Integrative Network Biology Analysis

PLOS Medicine Research Article

Comprehensive network biology analysis identifying key regulatory mechanisms in pancreatic cancer through integrative computational approaches.

Read Paper

/pet projects

Bayesian Flashcards project illustration

Bayesian Flashcards

A revolutionary study application that uses advanced Bayesian statistics to optimize your learning efficiency.

Code →
Adaptive Entropy-Gated Contrastive Fusion (AECF) project illustration

Adaptive Entropy-Gated Contrastive Fusion (AECF)

Robust, calibrated multimodal inference with dynamic entropy-gated fusion. Handles missing inputs in real-world systems.

Code →Paper →
AI Careers 2025 project illustration

AI Careers 2025

Interactive presentation exploring AI career paths, industry trends, and strategies for transitioning into artificial intelligence roles.

View Presentation →

/bayesian flashcards

GitHub Download DMG

Smarter Spaced Repetition, Powered by Bayesian Stats

Bayesian Flashcards is a next-gen study app that adapts to your learning curve. It uses Bayesian statistics to optimize review timing, so you remember more with less effort. Built for technical learners, ADHD brains, and anyone who wants to learn efficiently.
🔬
Adaptive Algorithm
Personalizes your revision schedule and difficulty using Bayesian inference.
🎨
Minimalist Design
Distraction-free, clean interface with code and math support.
⚡
ADHD-Friendly
Pomodoro timer, flexible sessions, and simple navigation for focus.
🔥
Motivation Mode
Keeps you in the sweet spot (70% accuracy) for optimal motivation.
📊
Progress Insights
Analytics highlight your strengths and what to review next.
🎓
Ready-to-Go Content
156+ coding interview questions included. Add your own easily.
My Vision: Make learning intuitive, effective, and enjoyable for everyone—whether you’re prepping for interviews or mastering new skills.

/contact

I’m open to freelance and contract work. Whether you’ve got a project or just want to chat, my inbox is always open.


✉️ Say hello