Experience¶

• Developed a semantic search engine that leverages historical production data, reducing by 70% the time needed to identify similar projects for a client in the industrial sector.
• Built a real-time tracking application for manufacturing components, ensuring complete traceability of the production flow and improving efficiency between workshops.
• Created and trained machine-learning models (one per work centre) to predict manufacturing times, reaching over 85% accuracy (MAPE < 15%) and making production planning more reliable.

• Semantic search
• Time prediction
• Manufacturing
• Python

Generative AI & data science

• Developed a multi-stage AI agent (LangGraph, GPT-4o-mini) to automate project initiation: it generates actionable tasks, identifies dependencies, builds the schedule, allocates resources, assesses risks and draws the project plan as detailed, automatically generated Gantt charts.
• Built QuizPhere, a full-stack quiz generation platform powered by generative AI (Groq, Llama 3.2): text quizzes plus relevant images found through DuckDuckGo Search, with True/False, Multiple Choice and Puzzle formats (LangChain, FastAPI, PostgreSQL, React).
• Developed a conversational AI for hackathon students with LangChain, the Groq API and Llama 3.1, using ChromaDB as a vector store, deployed with Docker behind REST endpoints.
• Set up a local LLM platform with Open WebUI and Ollama, plus Stable Diffusion for image generation.
• Created dashboards with Dash and Plotly for the company's games.

Computer vision

• Developed a face recognition Flask application that sends e-mail alerts, with screenshots, when someone enters a room without authorisation.
• Led a facial attribute analysis application: gender classification, age prediction and emotion analysis.
• Built an object counting by region application to monitor and manage assets in defined areas.
• Developed a real-time people detection and counting algorithm optimised for the NVIDIA Jetson Nano (edge computing).

• LangGraph
• LangChain
• RAG
• ChromaDB
• FastAPI
• Flask
• Docker
• React
• Jetson Nano
• Dash

• Implemented a deep learning approach for foot-type classification from heterogeneous pressure data.
• Designed the methodology to collect both image and numerical data.
• Fine-tuned VGG16 and InceptionV3 on the images, and K-NN and CART on the numerical data.
• Combined them with a stacking ensemble to improve the final model, then deployed it on OVH Cloud.

• Deep learning
• Transfer learning
• Stacking
• OVH Cloud

• Led a project to detect sources of infection in operating rooms.
• Built the image dataset and annotated it with Roboflow to create ground-truth labels.
• Trained YOLOv4 object-detection models for surgical instruments, mask wearing, medical caps and earrings.
• Deployed the models on the company's local servers.

• YOLOv4
• Object detection
• Roboflow
• Healthcare

• Developed an Excel application in VBA to manage inventory more efficiently and accurately.
• Implemented a machine-learning model to identify and reduce waste on the production line, and presented recommendations to improve production efficiency.

• Excel & VBA
• Machine learning
• Industry