My journey

Professional experience

A detailed look at my career path and achievements

May 2026 — Present
Montpellier, France · Freelance
Skermine Consulting is my own consulting company (SASU, Montpellier), through which I deliver AI engineering missions and AI training.
Independent since 2026: I design and teach production-grade agentic AI, and take on GenAI and LLMOps missions for companies in Switzerland and across Europe.
Key achievements
- Design and teach Simplon × Wild Code School's 6-month “Dev IA Agentique” bootcamp (RNCP37827) for developers moving into AI engineering
- Rebuilt the whole curriculum (phases 0–7) as competency-based modules: operational objectives, learning activities, assessment criteria
- Wrote the flagship briefs: a support agent with memory, guardrails and Langfuse evaluations from day one; a secure data gateway combining advanced RAG, Text-to-SQL and an MCP server
- Producing a French-language YouTube series that explains AI to professionals
Client engagements
Simplon
Tech training · RNCP37827 · 2026
Situation
Simplon and Wild Code School launched a 6-month agentic-AI developer bootcamp. The track needed a complete curriculum reflecting how AI systems are really built and run in production.
Task
Design the full curriculum (phases 0–7) and teach the 2026 cohort, with every competency taught and assessed through realistic professional situations.
Action
I rewrote every phase in a competency-based format, designed the flagship briefs (a support agent with memory, guardrails and evaluations; a secure RAG + Text-to-SQL + MCP data gateway), made a design phase mandatory before any code, and produced reference solutions with architecture diagrams.
Result
One consistent curriculum across all phases, taught to the 2026 cohort. Learners practice production habits — evaluations, observability, guardrails, design reviews — from their very first agent.
Technologies
Agentic AI · RAG · Text-to-SQL · MCP · Langfuse · Guardrails · Python · Curriculum design
Oct 2025 — Present
Remote · Part-time
ByteInnov is a multi-product tech company building digital and fintech products for the West-African (UEMOA) market.
I lead the data and AI strategy: scoping, prioritizing and de-risking AI initiatives for financial organizations in West Africa.
Key achievements
- Authored the AI scoping note: document AI (OCR, classification, semantic search), French–Fon speech & translation, MCP-based agents and explainable financial models
- Defined SMART targets, governance, staffing and budget, with market studies and POCs as go/no-go gates
- Benchmarked Whisper large-v3 against Fon-specific speech-recognition models on real voice notes
- Supported the leadership's portfolio arbitration across ten products ahead of a board meeting
Technologies
Document AI · OCR · NLP · Whisper · ASR · MCP · Python
Jan 2025 — Nov 2025
Paris, France · Full-time
Devoteam is a leading European consulting group that helps businesses leverage digital transformation through cloud, data, and AI solutions.
Delivered technical consulting for AWS-backed startups while contributing to the internal AI community through mentoring and knowledge sharing.
Key achievements
- Took ownership of 4+ startup engagements, designing and implementing production-grade ML and LLM solutions
- Conducted technical interviews and mentored junior engineers
- Led sessions on emerging AI trends like LLMOps and multi-cloud architectures
- Established best practices for LLM evaluation and multi-provider flexibility that became reusable assets
Client engagements
Kinetix
Audio AI Startup · 2025
Situation
Kinetix's GPU costs for Whisper inference were unsustainable at scale. Their NVIDIA-based infrastructure threatened unit economics as customer volume grew.
Task
Evaluate AWS Inferentia as a cost-effective alternative and migrate the Whisper inference stack without compromising transcription quality or latency.
Action
I ported the Whisper model to AWS Inferentia using the Neuron SDK, conducted rigorous benchmarks against NVIDIA Triton across various audio workloads, and implemented KV-Cache and I/O optimizations.
Result
Achieved 50% reduction in inference costs with a 10% performance improvement. Delivered a comprehensive benchmark report that informed Kinetix's long-term infrastructure strategy.
Start Catalog
Enterprise SaaS · 2025
Situation
The application was tightly coupled to OpenAI, creating vendor lock-in. Additionally, valuable enterprise knowledge was trapped in SharePoint repositories.
Task
Enable multi-provider LLM flexibility without disrupting existing functionality, and build a RAG pipeline to unlock SharePoint knowledge.
Action
I architected a feature-flip system for seamless OpenAI/Bedrock switching. Built a SharePoint crawling mechanism, indexed documents in OpenSearch with semantic chunking, and integrated a Bedrock agent with knowledge base.
Result
Delivered multi-provider LLM flexibility allowing runtime provider switching, plus a fully operational enterprise RAG system with secure, permission-aware document retrieval.
Welcome Account
FinTech Translation · 2025
Situation
The team had no industrialized way to measure LLM translation quality. Manual spot-checks couldn't scale, and regressions were only discovered after impacting customers.
Task
Build an automated LLM evaluation pipeline integrated with CI/CD, and implement observability for real-time monitoring of usage, latency, and costs.
Action
I implemented a DeepEval-based evaluation framework directly in the CI/CD pipeline. Added Sentry tracing for LLM calls and integrated Comet translation quality metrics.
Result
Established automated CI/CD LLM testing that catches regressions before deployment. Enabled live usage and cost monitoring giving leadership full visibility into AI spend.
Bevolta
Energy Tech Startup · 2025
Situation
AWS deprecated their Forecast service, forcing an urgent migration. Bevolta's system provided per-client forecasting models critical for personalization.
Task
Migrate from AWS Forecast to SageMaker AutoML while preserving per-client model personalization and implementing intelligent retraining.
Action
I designed a migration architecture using Autopilot with multiple algorithms. Built a DynamoDB-based performance tracking system with conditional retraining triggers. Created two operational modes: cost-conscious and exploratory.
Result
Achieved functional migration with zero service disruption. Delivered intelligent performance monitoring that triggers retraining only when needed, plus FinOps optimization reducing ongoing ML infrastructure costs.
Technologies
AWS Bedrock · AWS Inferentia · Neuron SDK · LangChain · RAG · DeepEval · Python · Terraform
Sep 2021 — Dec 2024
Paris, France · Full-time
Solocal is France's leading digital marketing company, connecting local businesses with customers through advertising and digital presence solutions.
Built and maintained production ML systems serving 4,000+ SMB clients, focusing on marketing attribution and real-time bidding optimization.
Key achievements
- Built a Bayesian MMM solution using Google's LightweightMMM, reducing ad spend waste by 15%
- Improved delivery accuracy from 60% to 85% through model optimization
- Designed Airflow pipelines on GCP Vertex AI for automated model retraining
- Optimized Flask API to handle 4M+ daily requests with 99.9% uptime
- Contributed to presales by translating ML capabilities into business value
Technologies
Python · GCP Vertex AI · Apache Airflow · Flask · Docker · Kubernetes · MLflow · PostgreSQL · Redis
Feb 2019 — Aug 2021
Paris, France · Full-time
Renault Digital is the innovation hub of Renault Group, driving the digital transformation of one of Europe's largest automotive manufacturers.
Developed predictive models and computer vision solutions for marketing optimization and manufacturing quality control.
Key achievements
- Developed Random Forest affinity models achieving 25% improvement in click-through rates
- Deployed customer preference models across France and Germany markets
- Fine-tuned Mask R-CNN for vehicle part detection in quality inspection
- Built automated dataset generation pipeline using Blender, accelerating CV iteration from weeks to days
Technologies
Python · TensorFlow · Mask R-CNN · OpenCV · Blender · GCP Vertex AI · BigQuery · Apache Spark