About Valentin
My Credo: I build production-grade AI systems that solve real business problems, moving beyond the MVP stage to deliver scalable impact.
As a Data Scientist with 8 years of experience (LVMH, Chanel, TF1, Club Med), I specialize in end-to-end AI leadership. I bridge the gap between business stakeholders and technical execution, ensuring that AI investments translate into operational reality.
My Areas of Expertise:
- Strategic Scoping & Business Alignment: Deep-diving into business needs to design AI solutions that people actually use.
- Robust AI Architectures: Designing complex LLM chains (e.g., automated Python code generation for custom triggers/email automation) with a focus on reliability.
- Advanced Evaluation (Evals): Implementing "LLM-as-a-judge" frameworks and continuous monitoring to ensure output quality and safety.
- Cloud & Orchestration: Building multi-source ETL and serverless deployments (GCP) optimized for cost and performance.
- Technical Coaching: Mentoring Data Science teams and tech profiles to elevate their production standards and architectural thinking.
Advanced Data Science & Deep Learning / ML Expertise:
- Recommendation & Similarity Engines: Developed product substitution systems and "Look Builders" using CNNs, NLP, and Transfer Learning for high-end retail and luxury brands.
- Deep Learning & Machine Learning : Proven track record in deploying custom architectures (Computer Vision, Advanced NLP) to solve specific business challenges where traditional AI and Deep Learning excel.
Founder of Artisan Business:
A production AI SaaS featuring multi-agent pipelines and RAG (GCP serverless, ~10k structured customer entries). I don't just advise; I build and operate.
French
Native or bilingual
English
Fluent
Can work on-site
Paris (up to 50km)
Experience
- Artisan BusinessFondateur & ingénieur IASOFTWARE PUBLISHINGNovember 2024 - Today (1 year and 7 months)FranceSaaS construit autour de pipelines IA pour aider les artisans d'art à gagner du temps sur leur admin
- Gestion avec les métiers : de l'idée au système en production en quelques semaines
- Pipelines LLM en production : orchestration multi-agents (segmentation, enrichissement, génération de messages) — 650 appels LLM / 30 jours, ~1,6 M tokens, fiabilité 99,7%
- Données unifiées : ingestion multi-sources (e-commerce, paiement, CRM, Gmail, Notion) via webhooks et batch, base Firestore — 10 utilisateurs bêta, ~10k entrées structurées
- Infrastructure serverless sur GCP : synchronisations asynchrones, fiabilisation continue, déploiement cloud
- Gouvernance & Fiabilité LLM : systèmes d’évaluation automatisés (LLM-as-a-judge), guardrails, etc pour monitorer la qualité des sorties, métriques de performance et systèmes auto-correcteurs en production
- Cycle produit complet : cadrage terrain avec les artisans, itérations rapides sur les parcours critiques
-----AI-driven SaaS helping art artisans automate their administrative workflows.- Production LLM Pipelines: Multi-agent orchestration (segmentation, data enrichment, message generation) — 650 LLM calls/30 days, ~1.6M tokens, 99.7% reliability
- Unified Data Architecture: Multi-source ingestion (E-commerce, Payments, CRM, Gmail, Notion) via webhooks and batch processing into Firestore — 10 beta users, ~10k structured entries
- AI Governance & Reliability: Implementing automated evaluation frameworks (LLM-as-a-judge) and safety guardrails to monitor output quality, track performance metrics, & self-correcting systems for robust production environments
- Serverless Infrastructure (GCP): Asynchronous synchronization, continuous reliability engineering, and cloud deployment for optimal scalability and cost
- Full Product Lifecycle: On-the-ground scoping with artisans, followed by rapid iterations on business-critical user journeys
Python LLM / LangChain LLM evaluations RAG GCP ETL Webhooks - DEMAIN - Cabinet conseilConsultant Data Science & Lead DataCONSULTING AND AUDITSMay 2018 - November 2024 (6 years and 6 months)Paris, France7 ans de conseil data chez des grands comptes (LVMH, CHANEL, TF1, M6, Club Med…), de la définition de stratégie jusqu’à la mise en production de systèmes ML complexes. Cadrage, roadmap & coaching des équipes.La mission dont je suis le plus fier : un pipeline de similarité visuelle (CNN) et textuelle pour LVMH, permettant de comparer et classifier des contenus visuels à grande échelle (use cases : produit de substitution et look builder) — de la recherche à la production en boutique et en ligne.Autres réalisations :
- Détection d’anomalies sur catalogue produit (prix, attributs, visuels) et optimisation référentiel
- Modèles CRM pour des clients e-commerce et médias : anti-churn, scoring, segmentation, attribution
- Analyse de performance moteur de recherche (TF1), segmentation CRM (M6, Club Med)
- Pilotage de la pratique data : stratégie, roadmap, accompagnement équipes clients
Clients : LVMH, Diptyque, CHANEL, TF1, M6, Club Med et autres grands comptes.Stack : Data Science · Vision/CNN · NLP · Vector DB · CRM · GCP · AWS · Snowflake · Databricks · Coaching · Agile--7 years of Data Consulting for major accounts (LVMH, CHANEL, TF1, M6, Club Med), managing everything from AI strategy to the production of complex ML systems. Scoping, roadmap design, and team coaching.Flagship Project: Visual & textual similarity pipeline (CNN) for LVMH, enabling large-scale content classification and comparison (Use cases: Product Substitution & Look Builder) — from R&D to live production (retail & online).Also :- Product Catalog Quality: Anomaly detection (price, attributes, visuals) and master data optimization
- CRM & Performance: Churn prediction, scoring, and attribution models for e-commerce and media leaders
- Strategic Leadership: Data practice management, roadmapping, and technical mentoring for client teams (TF1, M6, Club Med)
Skills: Data Science · Computer Vision (CNN) · NLP · Vector DB · CRM · GCP · AWS · Snowflake · Databricks · Coaching · Agile - CNRS-Assistant de recherche NLP - CNRS (DRM & CREST)RESEARCHJune 2017 - December 2017 (6 months)Paris, FranceNLP, modélisation (Topic modeling, LDA, similarité, sentiment analysis), production et analyse de données sur corpus FOMC et leurs liens avec des rapports financiers - Python
- Extraction & Préparation : Nettoyage de données textuelles (NLTK), gestion des stop-words métier et lemmatisation/radicalisation
- Classification & Topics : Identification automatique de sujets via LDA pour transformer des milliers de documents en thématiques actionnables
- Analyse de Sentiment : Scoring de polarité et de subjectivité (spécialisation en vocabulaire financier/économique)
- Mise en relation de documents : Analyse de similarité et détection de corrélations entre différentes sources de données (ex: rapports internes vs discours publics)
Stack technique : Python, Scikit-Learn, Gensim, NLTK, TextBlob, Pysentiment.
Recommendations
These freelancer profiles also match your criteria
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4
Education
- Ingénieur en Statistique & Science des DonnéesUniversité Paris-Dauphine2018Ingénieur en Statistique & Science des Données
- Classe préparatoire MP2015