Freelancer profile translated to English.
Back to original languageAbout Sarah
AI pipelines in production have one thing in common: they work in demos but collapse in production. I'm here to make sure that doesn't happen to you.
Senior AI Engineer, I design robust agentic systems, production-ready RAG pipelines, and multi-agent orchestrations for projects where complexity isn't a detail, but a reality. My clients: LexisNexis, Criteo, Eramet, environments where data is massive, unstructured, and critical: legal documents, industrial data, end-to-end automation workflows.
What I deliver:
✔ AI Agents and Multi-Agent Systems: LangGraph, Google ADK, MCP, for production-ready autonomous workflows.
✔ RAG Pipelines: indexing, chunking, reranking, evaluation, optimized to perform under real-world conditions.
✔ Multimodal Extraction: processing complex documents (PDF, legal, industrial, multilingual).
✔ LLM Evaluation Frameworks: metrics, ground truth, benchmarking, drift detection.
✔ Audit and Architecture: analysis and redesign of existing AI systems to make them resilient and scalable.
With a PhD in Computer Science, I combine academic rigor with field execution. No abandoned POCs: I build architectures that go into production and handle the load.
Available for freelance missions. Let's talk about your project.
French
Native or bilingual
English
Fluent
Arabic
Native or bilingual
Can work on-site
Paris (up to 50km), Marseille (up to 50km)
Experience
- CRITEOSenior AI EngineerE-COMMERCEOctober 2025 - Today (10 months)Paris, FranceContext:Criteo deploys large-scale AI systems and needed a rigorous framework to evaluate the reliability of its multi-agent pipelines in real-world conditions before any production deployment.Challenge:No standardized framework existed to objectively measure the performance of complex multi-agent systems: hallucinations, output consistency, behavior under load, agent drift.Solution:I designed a comprehensive multi-agent evaluation framework, including:
- Definition of adapted metrics
- Construction of ground truth datasets
- Automated benchmarking
- Continuous monitoring
The architecture integrates with existing CI/CD pipelines and triggers alerts in case of performance drift.Result:- Standardization of the LLM evaluation process at Criteo
- Early detection of regressions before production deployment
- Reusable framework for multiple product teams
- ErametSenior AI EngineerRAW MATERIALS INDUSTRYJune 2024 - September 2025 (1 year and 3 months)Paris, FranceContext:Eramet, an international mining group, produces massive volumes of unstructured geological reports. Manual data extraction was slow, costly, and error-prone.Challenge:Automate the analysis of complex technical documents (geological PDFs, estimation reports) and make the data usable by business teams, without systematic human intervention.Solution:I designed and deployed a production-grade RAG pipeline including:
- Multimodal extraction via Azure Document Intelligence
- Optimized indexing with Azure AI Search
- CSV Agents to handle volumes exceeding LLM context windows
- Complete Prompt Flow workflow, with CI/CD and N8N integration for geological report automation (GPT-4)
Result:- Full automation of the geological extraction and reporting cycle
- Significant reduction in manual processing time
- Pipeline deployed in production in an Agile environment, with iterative deliveries
- Saint-GobainSenior Data EngineerRAW MATERIALS INDUSTRYJune 2023 - June 2024 (1 year)Courbevoie, FranceDATA4LCA:Automation project for the Life Cycle Assessment (LCA) of Saint-Gobain products, with plasterboard as the case study.Mission Activities:
- Understanding and analysis of business needs
- Identification of data requirements
- Data exploration with Python
- Definition and monitoring of data quality
- Automation of LCA data collection
High Performance Solutions (HPS):Data migration, ingestion, and processing projectMission Activities:- BW to Synapse migration
- Data ingestion into Snowflake
- Ingestion of data from ERPs (QAD, SAGE) into Synapse
- API data ingestion
- Data ingestion and processing with Spark/Python
- CI/CD
- Agile environment (Scrum)
Reviews
Recommendations
JB
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
- PhD in Computer Science, Anonymization of Large DatasetsUniversité Sorbonne Paris Nord2020
- State Engineer Diploma in Operations Research and Decision SupportINSEA2014
Certifications
- Microsoft Certified: Azure Data Scientist AssociateMicrosoft2022
- Databricks Certified Data Engineer AssociateDatabricks2023