Daniel Perez
Senior AI & Data Scientist | LLMs, Fraud Detection
Salaire / Taux journalier
Paris, France
Freelance
Peut venir sur site
Expertises
Langues
À propos
Introduction
AI & Data Scientist with PhD from ENS, specializing in LLM applications and fraud detection systems. Led team of 8 at Qantev, reducing false positives by 50% and building production-grade AI pipelines for AXA and BUPA. Founder of Hexagone Leads, a GDPR-compliant B2B prospecting SaaS. Deliver end-to-end AI solutions—from LLM document processing to fraud scoring—helping insurance and B2B companies automate operations and scale intelligently.
Expérience de travail
Founder
Hexagone Leads
sept. 2025 - Présent
1 an 2 mois
Paris, France
Built Hexagone Leads from 0→1: a B2B prospecting SaaS combining French official legal data, financial signals, and in-house Al to surface qualified leads in minutes 100% GDPR compliant. Technical highlights: – Architected an Al-powered ICP matching engine using LLMs and vector search on enriched company profiles – Integrated multi-source data pipelines into a unified enrichment layer – Designed a recommendation system matching user-described products to relevant companies across a 26M+ entity database – Full product ownership: ML pipeline, frontend/backend architecture, and go-to-market
Independent Consultant
Hexagone Labs
sept. 2025 - Présent
1 an 2 mois
Paris, France
Al & Data freelance practice operating under Hexagone Labs. Selected engagements: – Designed and deployed LLM-based document processing pipelines for insurance clients, significantly reducing manual review time – Built end-to-end fraud scoring systems integrating claims data, network analysis, and behavioral features – Advised on data infrastructure and ML model productionization for B2B SaaS startups – Developed Al-powered automation workflows for sales teams: lead enrichment, ICP matching, outreach sequencing – Delivered data strategy roadmaps for companies transitioning to Al-first operations
Qantev
juin 2024 - juin 2025
1 an 1 mois
Paris, France
Fraud, Waste and Abuse - Al Team Lead
janv. 2025 - juin 2025
6 mois
Paris, France
Led the Al team focused on Fraud, Waste & Abuse (FWA) detection for health insurance claims, serving clients across APAC, EMEA, and LATAM. Handled clients of international handling like AXA and BUPA. – Managed a team of 8 data scientists across the full model development and production deployment lifecycle – Reduced false positives by 50% through advanced anomaly detection and outlier scoring models on high-volume claims data – Spearheaded the development of LLM-powered agents using LangGraph to identify emerging fraud patterns automatically within client data – Built LLM pipelines using Azure-hosted GPT models (Azure Al Foundry) for document analysis triage – Optimised R&D processes and investigation workflows, boosting operational efficiency by 25% – Built model explainability and auditability layers to meet insurance compliance requirements across multiple regulatory jurisdictions
Data Scientist
juin 2024 - janv. 2025
8 mois
Paris, France
Developed ML models for health insurance claims analysis: anomaly detection, provider profiling, fraud scoring, improving operational efficiency by 40% – Built the company's document fraud detection solution, automating the identification of fraudulent documents within insurance claims – Built and annotated training datasets using Active Learning for supervised fraud classification tasks – Constructed feature engineering pipelines on structured and semi-structured claims data – Benchmarked gradient boosting models (CatBoost, XGBoost, LightGBM) and deep learning models (PyTorch) against production baselines Promoted to Al Team Lead within 8 months
Research and Development Project Manager
Eco-Adapt
janv. 2023 - juin 2024
1 an 6 mois
Paris, France
Deployed ML solutions for predictive maintenance of industrial equipment at Eco-Adapt, a subsidiary of Schaeffler, a global industrial group. – Developed new product features using cutting-edge signal processing techniques, reducing uncertainty by up to 30% in predictive maintenance systems – Designed anomaly detection algorithms on sensor time-series data, reducing false positive rates and introducing confidence bands – Benchmarked models against industry baselines — consistently outperformed on both real-world and test data – Coordinated with industrial clients to align model outputs directly with maintenance scheduling decisions – Contributed to the Predict-Adapt platform used by Schaeffler's industrial client base
Formation
Ecole normale supérieure
PhD Candidate, Mathematics
2019 - 2023
3 ans 4 mois
– PhD in Mathematics. "Persistent homology of stochastic processes and their zeta functions", supervised by Pierre Pansu and Claude Viterbo (ENS Ulm / Université Paris-Saclay). – Proved stochastic Wasserstein stability of persistence barcodes: a result directly applicable to ML model robustness under distributional shift – Introduced homological indices and dimensions linking topological data analysis to geometric and probabilistic structures – Designed statistical discrimination tests for stochastic processes, applicable to time-series anomaly detection and signal classification – Introduced ζ-functions for stochastic processes and computed them for Brownian motion and α-stable Lévy processes – Published 3 papers (2020, 2020, 2021); visiting researcher at the University of Chicago (hosted by Shmuel Weinberger, 2022) – Doctoral contract MESRI Key transferable skills: probabilistic modeling, topological feature extraction, distributional testing, mathematical rigor
Université Paris-Saclay
Teaching Assistant, Mathematics
2019 - 2021
2 ans 4 mois
Teaching assistant for mathematics courses at Université Paris-Saclay, alongside PhD research. – Mathematics of Modelling (M191) – Linear Algebra for Physicists (M259) – Measure Theory (M302) – Linear Algebra 2 (M252)
Laboratoire Mathématique d'Orsay
Intern, Mathematics
2019 - 2019
4 mois
Research in Topological Data Analysis persistence modules, field dependence of persistent homology, and geometric applications to point cloud data.
Université Paris-Saclay
Master of Science - MS, Mathematics
2018 - 2019
11 mois
Centre de physique théorique
Intern, Theoretical Physics
2018 - 2018
5 mois
Formalized physics using differential geometry and representation theory. Derived Dirac equations geometrically; studied Einstein-Cartan gravity and spontaneous symmetry breaking.
École Polytechnique
Master of Science - MS, High Energy and Particle Physics
2017 - 2018
1 an 1 mois
Laboratoire de Physique Théorique et Hautes Énergies
Intern, Theoretical Physics
2017 - 2017
4 mois
Multi-loop diagram methods in quantum field theory. Computed β-function and critical exponents of the q^4-model to two loops.
Pierre and Marie Curie University
Master, Theoretical and Mathematical Physics
2016 - 2017
1 an 1 mois
University of Ottawa
Bachelor of Science (BSc), Theoretical and Mathematical Physics
2013 - 2017
4 ans 1 mois
Licences et certifications
CCENT
Délivré le : –
Portfolio
