Francois Courbier

Engineered automated recommendation systems for corrections and adjustments on multi-dimensional incoming real estate

• • data (textual and numerical) delivered via high-volume Excel files. R&D: Conceived a hybrid machine learning model (Graph Theory + NLP) and conducted a comparative evaluation against

◦ ◦ GPT-4o to quantify and minimize hallucination rates. Prod: Developed Python backend microservices deployed via Docker, and performed hardware infrastructure sizing studies tailored for Big Data workloads (20k Excel files containing up to 20k rows each), yielding an 80% time-saving across processed files. Identified, mapped, and prioritized corporate AI use cases leveraging internal company data. Delivered AI awareness and training sessions for internal teams, with a specific focus on compliance with the EU AI Act.

Architected real-time enrichment pipelines for a continuous stream of customer reviews (comments and ratings, ~1k entries

• • daily, millions in the database).

◦ ◦ R&D: Designed machine learning models covering sentiment analysis, NLP classification (rule-based & deep learning, up to 90% accuracy), extractive statistical text summarization (evaluated via the ROUGE metric), as well as anomaly and weak signal detection (Fourier transforms). Prod: Built and deployed production-grade Python backend microservices, implemented automated model retraining workflows, and established synchronous/asynchronous distributed computing architectures. Collaborated on benchmarking and fine-tuning open-source and proprietary LLMs (Llama 2, Dolly, ChatGPT 3.5) to automate response generation for social media commentary. Developed a Proof of Concept (PoC) for image analysis (hospitality/rental properties) using Convolutional Neural Networks (CNN) to detect anomalies and property issues (e.g., pests, water damage).

Steered a cyber risk assessment and modeling project targeted at space system accreditation.

• •

• • Coordinated international cross-functional teams adhering to Agile methodologies.