Freelancer profile translated to English.

Description

I help tech teams move from AI prototypes to production-ready systems, at scale.

10+ years as Staff / Lead Applied Scientist at Zalando and Trade Republic. My work blocks €1.5M in fraud per month at Trade Republic and generated €7M in annual EBIT at Zalando through size recommendation systems. Two publications at RecSys (2020, 2022).

How I can help you

AI Architecture & Strategy: System reviews, LLM/ML roadmaps, build-vs-buy trade-offs. This is where I bring the most value.

Senior Implementation: LLM fine-tuning, RAG, recommendation, vision-language, real-time ML.
Production Deployment: Robust pipelines designed for tens of millions of users (AWS, PyTorch, SageMaker).

What sets me apart:

Delivered products at Zalando's scale, not demos. Real production ML experience: latency, A/B tests, on-call. Deep expertise in multimodal (diffusion models, 12M image datasets) and recommendation (SASRec, bandits, two RecSys papers).

What I work on:

Audit or redesign of existing AI systems
LLM / ML roadmaps for founders and CTOs
Recommendation systems (e-commerce, content, products)
Production LLM pipelines: RAG, fine-tuning, evaluation
Fraud detection and real-time scoring
Logistics optimization and operational research

Ideal for founders, CTOs, and tech leads looking for a senior partner to design, audit, or build their AI systems. An architecture review or a short scoping mission is often the best way to start.

Industry field of expertise

Languages

English
Fluent
French
Fluent
German
Fluent

Workplace preferences

Can work on-site

Paris (up to 50km)

Trade Republic
Staff Data Scientist — FinCrime
BANKING AND INSURANCE
August 2025 - Today (1 year)
Paris, France
I joined Trade Republic to structure the ML strategy for the FinCrime team and deploy a real-time fraud detection model.

Defined the team's technical ML strategy:
- Data and labeling roadmap, platform direction, modeling approach.
- Deployed a low-latency Transformer model that blocks €1.5M in outgoing APP fraud per month.
- Refactored ML tooling to standardize the train/register/deploy workflow and accelerate iterations.
Pytorch Clickhouse AWS Apache Kafka DBT
Zalando
Lead Applied Scientist — Size and Fit
E-COMMERCE
August 2022 - August 2025 (3 years)
Berlin, Germany
I led the research and delivery of Zalando's size recommendation systems, on products used by tens of millions of customers.

- Led "Your Sizes", Zalando's size recommendation system.

- Reduced size-related returns by 0.2 percentage points, representing approximately €7M in annual EBIT.

- Led research on size-aware Virtual Try-On: fine-tuning diffusion models Leffa and CogView 4 to preserve garment fit and body shape, and fine-tuning Janus7B to annotate a 12M image dataset.

- Adapted Llama 3.1 (LoRA) to personalize size recommendations.

- Deployed a 3D body prediction model (ConvNet), a shape inference model (U-Net denoiser), and a multimodal Transformer for size prediction.

- Defined the technical roadmap, managed intake, and implemented evaluation frameworks (manual and automatic).
Stable Diffusion LoRa Pytorch AWS Databricks