Freelancer profile translated to English.
Back to original languageAbout Clément
Forward Deployed Engineer (FDE) | GenAI in production
You have an AI use case that has taken off internally — a conversational agent on your business data, an assistant for your documentation base, a workflow automation. It has become useful, sometimes critical. The next step is to put it into production: no one on the team has the time or the experience.
This is the job of a **Forward Deployed Engineer**: I embed myself within your teams, translate your business problem into an AI system, and build it in your infrastructure — from POC to MVP to production. Not a demo: code that runs at your place, under your constraints (backend, **API integration**, LLM orchestration, agents).
And above all, your AI becomes reliable:LLM observability (LLMOps)tells you when it deviates, and why.
At Club Med, for nearly2 years**, I laid the foundations for their generative AI in production: a RAG + **multi-agentconversational system, with:
—Langfuseobservability to trace every reasoning step and detect deviations;
—non-regression testsscored by a second LLM (LLM-as-judge);
— a multi-model gateway to switch models on the fly (OpenAI, Anthropic, Mistral…) without touching the code, and route each request to the cheapest suitable model —−50%inference costs.
Depending on where you are, three levels:
1. **Audit**: Is your AI ready for production? You leave with a diagnosis and a quantified plan.
2. **Industrialization**: Full production deployment — observability, evaluations, CI/CD, environment isolation, AI code security.
3. **Part-time FDE**: A few days a month to evolve it and keep it reliable.
You keep an embedded specialist without the cost of a full-time employee.
An AI use case to put into production? Tell me where you are in three lines. I'll reply within the hour.
French
Native or bilingual
English
Native or bilingual
Can work on-site
Paris (up to 50km), Bordeaux (up to 50km), Lyon (up to 50km), Nantes (up to 50km), Marseille (up to 50km)
Experience
- Plateforme de l'inclusion
On Malt
AI Engineer | LLMOps, Agents & EvalPUBLIC SECTORJune 2026 - Today (1 month)Paris, France🎯Mission ObjectiveAudit and industrialization, as an embedded AI engineer (Forward Deployed Engineer), of Autometa — the internal GenAI conversational assistant for GIP Plateforme de l'Inclusion (beta.gouv.fr) that transforms natural language questions into data, reports, and dashboards from Matomo and Metabase. Goal: to move a promising POC into reliable production — DevOps foundation, AI dashboard quality, and agentic system robustness.🔧Achievements- Audit of the technical setup and AI dashboard quality, bug fixes, and DevOps foundation definition (automated deployment, dev/test/prod environment isolation, hosting).
- Industrialization of the GenAI layer: multi-model switching (Anthropic Claude ↔ Ollama Cloud ↔ Scaleway GenAI API) by endpoint redirection, without touching application code, to balance cost/quality.
🧰Technical Stack- **Backend**: Python 3.14, FastAPI, Uvicorn, SQLAlchemy, Alembic, Jinja2
- **GenAI / LLMOps**: Claude (Claude Code in agentic mode), Ollama Cloud, RAG, skills/agents, LLM-as-judge evaluations
- **Data**: PostgreSQL, Redis, Matomo, Metabase
- **Frontend**: htmx, Bootstrap, Mermaid
- **DevOps / Infra**: GitHub Actions, Scalingo (PaaS), Scaleway (Object Storage, Functions), Docker, oauth2-proxy
- **Quality / Security**: pytest, Ruff, Bandit, Gitleaks, Sentry, OpenTelemetry
- Club MedAI Engineer | RAG, Agents & LLMOpsENTERTAINMENT AND LEISUREMay 2023 - June 2025 (2 years and 1 month)Paris, France🎯 Mission ObjectiveDesign and production deployment, over 2 years, of a generative AI backend platform for Club Med's Global Marketing Digital & Technology: document ingestion, full RAG, multi-agent orchestration, multi-model LLM gateway (LiteLLM), data exposure via internal APIs, and automation of a scalable cloud infrastructure with a FinOps approach.🔧 Achievements
- Design and production deployment of acomplete generative AI platform**: document ingestion, full RAG, multi-agent orchestration, and multi-model LLM gateway via **LiteLLM—2 years in productionserving Global Marketing teams.
- LLM monitoring and observability (**Langfuse**): tracking costs, latencies, generation quality, and detecting production deviations.
- Non-regression tests and automated evaluation of LLM responses (Promptfoo, LLM-as-judge approach): a second model scores outputs to ensure quality with each system update.
- Development ofPython / FastAPIAPIs to provide self-service data to business teams — containerized backend services, continuously deployed on the cloud.
- FinOps strategy for the GCP scope:-50% cloud costswithout service degradation, including LLM inference cost optimization (caching, batching, task-calibrated models).
- Cloud infrastructure automation in IaC (Terraform) and CI/CD pipeline industrialization.
🧰 Technical Stack- **Backend**: Python (FastAPI, uv, Streamlit), SQL, Bash
- **Generative AI**: LiteLLM, LangChain, ChromaDB, Vertex AI, Flowise, Langfuse
- **Cloud & Data**: GCP (BigQuery, Dataform, Cloud Run, Vertex AI, Workflows, CloudSQL)
- **Containerization**: Docker, Docker Compose, Ansible
- **IaC & CI/CD**: Terraform, GitLab, GitHub
- Livetrend.coCloud Engineer | Multi-Cloud GCP/AWS & PostgreSQLFASHION AND COSMETICSMarch 2025 - May 2025 (2 months)Paris, France🎯 Mission ObjectiveEvaluate AlloyDB (GCP) as a cost-effective alternative to Aurora (AWS) for Livetrend.co (B2B mode) by replicating a realistic multi-cloud production configuration, with a redesign of inter-cloud connectivity and adaptation of backend jobs to ensure fair comparisons.🔧 Achievements
- Operational inter-cloud network architecture (high availability VPN + BGP routing) to secure backend traffic between GCP and AWS
- Deployment of a managed AlloyDB cluster equivalent to Aurora for benchmarking with identical configuration
- Adaptation of backend ETL jobs (orchestrated via Prefect) to run in the new VPC + interaction with AlloyDB
- Resolution of backend network issues (DNS, inter-VPC mounting, EFS replication) to ensure data ingestion reliability
- Validation ofAlloyDB for productionwith better performance/cost ratio + identification of blocking points for large-scale migration
🧰 Technical Stack- **Data Backend**: PostgreSQL (AlloyDB, Aurora)
- **GCP Cloud**: AlloyDB, Cloud VPN, Cloud Router, VPC
- **AWS Cloud**: ECS, EFS, Aurora PostgreSQL
- **Network**: HA-VPN, BGP, Site-to-site VPN, VPC Peering
- **Orchestration**: Prefect
Reviews
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
- Engineering Degree, GeneralistEcole Centrale de Marseille2022Diplôme d'ingénieur, Généraliste