Data Science & Machine Learning

Models that ship, not just notebooks

Twenty years building production systems, now applied to machine learning. Eleven projects covering supervised learning, computer vision, MLOps, RAG and LLM evaluation — each one taken from raw data through to a deployed, monitored, tested service.

OpenClassrooms — Data Scientist / Machine Learning (RNCP level 7, Bac+5)

Ask my AI about these projects

It answers from the same knowledge base these projects were written from.

Projects

City skyline of office towers at dusk with lit windows

Data Analysis & Supervised Learning

Exploratory analysis and supervised learning models — from market-expansion statistics to regression and classification — each one taken from raw data to a validated model.

Exploratory Analysis of Education Systems Data
Built a reproducible Jupyter environment with docker-compose and Poetry-managed dependencies.
Building Energy Consumption Forecasting
Full EDA and feature engineering, including using the proportion of energy types as a feature — which markedly improved results.
HR Attrition Classification
Trained and compared a dummy baseline, a linear model and a non-linear model on both train and test sets.

PythonPandasNumPyJupyterDockerPoetryscikit-learnMatplotlib

MLOps & Deployment

Trained models shipped as tested, monitored production services, with experiment tracking, CI/CD and drift detection.

Deploying a Machine Learning Model
Exposed the model through a FastAPI service, with Pydantic validating every incoming payload.
Credit Scoring — Experiment Tracking with MLflow
Tracked every run in MLflow — hyperparameters, metrics and model versions — so results stayed reproducible and comparable.
Credit Scoring — Deployment, Monitoring & Drift DetectionView code
Served the model through a FastAPI endpoint and built a Streamlit dashboard explaining why each credit decision was made.

FastAPIPydanticPostgreSQLPytestDockerHugging Face SpacesPythonMLflowLightGBMscikit-learnGridSearchCVStreamlitCI/CDEvidently

Medical scan images displayed on a light box

Computer Vision

A semi-supervised image pipeline combining pretrained embeddings with clustering to label medical imaging data with minimal manual labelling.

Semi-Supervised Image Labelling
Extracted visual embeddings with a pretrained ResNet backbone, turning raw images into exploitable feature vectors.

PythonPyTorchResNetPCAscikit-learn

Visitors viewing paintings in an art gallery

LLMs & RAG

A deployed Retrieval-Augmented Generation chatbot together with its own rigorous evaluation, from corpus indexing to measured answer quality.

RAG System for Cultural Event RecommendationsView code
Built the corpus from live cultural event data pulled via the Open Agenda API, then cleaned, chunked and structured it for indexing.
Evaluating an LLM with RAGASView code
Selected and justified RAGAS metrics — faithfulness, answer relevancy, context precision and context recall — rather than reporting whatever the library emitted.

LangChainFaissMistral AIFastAPIPytestRAGASPydanticPython

A team planning session with sticky notes on a glass wall

AI Strategy & Recommenders

AI project leadership and a full recommender system — from scoping and architecture to a working, deployed application.

Framing an AI Project — Lead Role
Scoped a PoC around the core product feature — matching images of clothing — and ran a technology watch comparing recent computer-vision models against classical approaches.
Crop Recommendation System for AgricultureView code
Merged two separate crop-yield datasets into a single enriched dataset, then built the full pipeline: preprocessing, hyperparameter optimisation, evaluation.

System DesignComputer VisionCloud ArchitecturePoCPythonMLflowStreamlitscikit-learnPandas

Questions about any of this?

Ask about the modelling choices, the trade-offs, the deployment, or anything on my CV. My AI answers in your language.

Start a conversation