ML Developer & Analyst

Alex Webb

Geospatial pipelines, probabilistic reasoning, quantitative risk modelling. Interested in how intelligent systems fail. Based in Lausanne, Switzerland.

LinkedIn / GitHub / alexwebb@pm.me

About

Senior developer with an end-to-end approach: from raw data and model architecture through to deployed applications. Background spans geospatial ML, satellite imagery analysis, quantitative risk tooling, and full-stack development.

Recent work includes SAR-based flood detection pipelines built and deployed at UNOSAT (UN Satellite Centre), and a suite of quantitative risk tools expressing a live macro thesis on energy transition. Open to contracts in ML engineering, geospatial analysis, and quantitative modelling.

Projects

InferSAR

PyTorch · UNet · GDAL · SAR · Python

Lightweight Python toolkit for on-device in-field flood inference from TerraSAR-X imagery. Cross-platform, no GPU required. Evolved from FloodAI v2, a UNet pipeline built during a contract at UNOSAT and used in UN disaster response workflows.

GitHub →

Quantitative Risk Suite

Python · Pandas · NumPy · Matplotlib

Portfolio-level risk tooling built around an energy transition macro thesis — FX gating, volatility-derived stop systems, correlation analysis, and rebalancing tools. Used on a live portfolio with real capital. Markov regime detection in development.

GitHub →

AYOU — Cooperative LLM Network

Python · RAG · LLM · Flask · Vector DB

Network of LLM agents with retrieval-augmented memory. Each agent maintains its own knowledge store; a routing function determines whether to answer locally or delegate to another agent. Built in 2022 before RAG architectures were mainstream.

Live → GitHub →

Transformer Attention Visualiser

PyTorch · Transformers · JavaScript · Python

Interactive visualisation of attention weight distributions across transformer heads. Illustrates how positional and semantic relationships are encoded — useful for understanding model behaviour and alignment failure modes.

GitHub →

Probabilistic Gene Trait Inference

Python · Bayesian Networks · Pandas

Given a CSV of related individuals and observable traits, infers probability distributions over genotypes and phenotypes using a Bayesian network. Applied to GJB2 hereditary hearing loss data.

GitHub →

Reinforcement Learning Othello

Python · Q-Learning · Flask · JavaScript

Othello engine trained via Q-learning through approximately 10,000 games of self-play, No hard-coded strategy — learns entirely from experience. State-action values updated via the Bellman equation. Playable in browser.

Play → GitHub →

Minimax Noughts & Crosses

Python · Minimax · Alpha-Beta Pruning · Flask

Tic-tac-toe engine implementing minimax with alpha-beta pruning — unbeatable by construction. Contrasts with the Othello engine: rule-derived optimality versus learned policy from experience.

Play →

Skills

ML & Data

PyTorch / TensorFlow
CNNs / UNet / Transformers
Bayesian inference
Reinforcement learning
Pandas / NumPy
Scikit-learn

Geospatial

SAR imagery analysis
GDAL / rioxarray
STAC / dask
QGIS
Sentinel-1 / TerraSAR-X

Engineering

Python / JavaScript
Flask / Django / React
Docker / CI-CD
Ubuntu / Nginx
SQL / REST APIs
Git / GitHub