Machine Learning
Understanding Isolation Forest for Anomaly Detection
Isolation Forest is one of those machine learning algorithms that seems almost too simple, yet surprisingly powerful.
Oct 7, 2025
01
About Me
I'm a Data Scientist, Data Engineer, and AI Engineer - the kind of profile that bridges the gap between research and production.
Over the past few years, I've worked across the full AI stack: designing Big Data pipelines that process 500M+ time-series data points at SNCF, building hybrid recommendation systems combining knowledge graphs and NLP, and shipping LLM-powered agents and RAG pipelines for real clients as a freelance AI developer.
My current focus is Generative AI engineering - RAG architectures, multi-agent systems with LangGraph, and LLM deployment. I care about systems that actually work in production, not just in notebooks.
When I'm not building, I write about ML on Medium and contribute to open-source AI projects on GitHub.
Tech Stack
Python / PyTorchTensorFlow / KerasNextJS / FastAPILangChain / LangGraphScikit-learn / XGBoostDocker / KubernetesAzure / GCPCelery / Redis / n8n
Education
Master's in Data Science
Université Lumière - Lyon, France
Bachelor's in Data Science
Université Lumière - Lyon, France
Bachelor's in Software Engineering
IFRI - Calavi, Benin
Soft Skills
AdaptabilityTeam PlayerTechnical CuriosityRigor & PrecisionProblem Solving
Languages
French
Native
English
Professional B2 Level
Certifications
Python - Kaggle
Machine Learning Engineer - Google Skills
GCP Essentials - Google Skills
02
Work Experience
AI Developer (Freelance)
Freelance
November 2025 - Present
- •Built and shipped production-grade AI systems for independent clients - from architecture to deployment.
- •Designed and deployed autonomous multi-agent systems (LangGraph orchestration, memory, tool calling) for 3+ clients in production.
- •Built end-to-end RAG pipelines (LangChain, HuggingFace, vector store) integrated into FastAPI web applications.
- •Shipped Kontact AI (mykontact.ai) - a LinkedIn prospecting SaaS with LLM scoring, scraping pipeline, and automated outreach sequences.
- •Reduced manual processing time by ~60% through inter-tool workflow automation (APIs REST, Make, Airtable).
PythonNextJSFastAPILangChainLangGraphSupabaseUnipileOpenRouter AIDockerHuggingFace
Data Scientist / Data Engineer (Internship)
SNCF Réseau
March 2025 - September 2025
- •Designed and industrialized Big Data pipelines (Spark, Databricks) for anomaly detection across 500M+ time-series data points in a railway context (level crossings).
- •Selected and implemented algorithms (LSTM Denoising Autoencoder, Isolation Forest, Transformers) with comparative performance analysis (precision/recall) and adaptation to business noise.
- •Optimized processing: 40% reduction in computation time and 30% improvement in flow reliability via business sequencing and tailored preprocessing.
- •Production monitoring with concept drift detection and business alerts (MLOps via MLflow).
- •Supported operational teams in defining use cases and achieved a 25% reduction in false positives through contextual modeling.
PythonAzure DatabricksApache SparkTensorFlowKerasScikit-learnPower BIMLOps
R&D Data Engineer (Internship)
EffetB
April 2024 - August 2024
- •Built a hybrid recommendation system combining Neo4j knowledge graphs and advanced NLP - reducing cold start by 30% through contextual embeddings (TF-IDF, Keras).
- •Applied advanced graph algorithms (GraphSAGE, FastRP) and dimensionality reduction (t-SNE) with 3D visualizations for analysis and explainability.
- •Validated the POC for product integration and concrete user-facing use cases.
PythonNeo4jScikit-learnNLTKKerasGraphSAGEFastRPt-SNE
Data Scientist
Intside LLC
December 2021 - August 2023
- •Built a KMeans clustering model identifying 4 strategic customer segments and lead scoring models integrated into the agency's targeting campaigns.
- •Delivered Power BI dashboards and collaborated cross-functionally with product, tech, and marketing teams.
PythonScikit-learnPower BISQLGCPKMeans
Software Developer / Data Analyst
Alibora SARL
December 2021 - August 2022
- •Built a call-center management system, contributed to ETL pipelines, and developed REST APIs with Laravel.
DockerJavaScriptLaravelSQLAPI Development
03
Skills & Technologies
Languages
PythonSQLRTypeScript
3+
Years Experience
15+
Projects Done
30+
Technologies
04
Featured Projects
Kontact AI
A production LLM-powered SaaS for LinkedIn prospecting - AI scoring, automated scraping pipeline, and outreach sequencing at scale.
Agent IAOpenRouteur AIAutomatisationSupabasePythonNextJS
Emotional Report
Voice-based emotion detection system using Wav2Vec2 for audio feature extraction and LSTM networks for multi-class emotion classification.
PythonWav2Vec2LSTMStreamlitDeep Learning
More Projects
LisboMercato
E-commerce platform with recommendation engine using collaborative filtering.
Next.jsTypeScriptAPI REST
Anomaly Detection Pipeline
Production-grade anomaly detection pipeline on 500M+ time-series records. LSTM Autoencoder for pattern learning combined with Isolation Forest for outlier scoring - deployed on Azure Databricks with PySpark.
PythonTensorFlowPySparkAzure
Knowledge Graph Recommender
Hybrid recommendation engine combining Neo4j knowledge graphs with NLP embeddings - 30% cold start reduction via contextual embeddings and community detection (GraphSAGE, FastRP).
Neo4jGraphSagePythonKeras
RAG Document Assistant
End-to-end RAG system for intelligent document Q&A. Hybrid retrieval (dense + BM25), re-ranking, and grounding check to minimize hallucinations. Built with LangChain, FastAPI, and a vector database.
LangChainOpenAIVector DBFastAPI
06