Pierre Allain

– Designed and trained transformer-based models on large-scale text corpora, covering data curation, domain-specific fine-tuning, and structured evaluation, with outputs deployed into production research and live trading systems. – Built a benchmarking framework for LLMs and encoder models on financial text, formalising task definitions and evaluation protocols to enable rigorous model comparison. – Built and optimised large-scale text retrieval over 10B+ documents using Elasticsearch, enabling efficient exploration. – Designed and deployed data pipelines to curate, enrich, and validate 20+ large-scale datasets used in research. – Mentored interns on applied machine learning and reproducible NLP workflows.

– Built Checkout.com's first machine-learning fraud detection system on 400M+ credit card transactions; designed a feature pipeline combining payment metadata, device intelligence, network signals, and velocity features. – Delivered the model as a low-latency scoring service integrated into the real-time authorisation flow, serving millions of predictions weekly under strict latency constraints. – Led follow-on modelling initiatives, including a payment traffic forecasting project and cross-team ML collaborations. – Mentored a senior product manager transitioning into a technical role through a one-year data science programme.

– Developed automated trading signals for US equities using fundamental data and machine learning, delivering 4 signals end-to-end with live Sharpe ratios above 2. – Built reusable ML pipelines for feature engineering, backtesting and reporting used across the research team.

– Built predictive models for PJM electricity prices using regression, time-series modelling, and simulation methods. – Developed and maintained live data pipelines used by traders and quantitative analysts.