Theo CREPIN

Development of a complete service for the extraction and verification of aeronautical release and maintenance documents. • Implementation of a data extraction pipeline from PDFs and photos using OCR and document understanding models. • Information matching and comparison to validate the maintenance status of parts and aircraft from multiple formats. • Fine-tuning and deployment of models on Vertex AI to improve extraction accuracy. • Integration of a specialized LLM trained on aeronautical terminology to analyze and confirm document compatibility. • Development of a global API (FastAPI) for automated document upload, extraction, and verification. 💻: Python, FastAPI, Vertex AI, OCR, LLM.

Development of a complete mobile solution for automatic office inventory. • Creation, annotation and processing of a dataset for office furniture detection from videos. • Use of transfer learning methods on YOLO and RF-DETR models and development of an object deduplication pipeline. • Development of a mobile application for client site visualization and room scanning. • Deployment of the furniture counting pipeline on RunPod and connection to the application and back-end database. 💻: Python, OpenCV, Docker, React-Native.

In the context of research projects, I developed two computer vision and NLP products trained on a Domino cloud. • Grouping, analysis and processing of a complex DICOM dataset. • Contribution to a project evaluating arthritis severity from foot and hand scans using PyTorch and OpenCV for subject selection in pre-clinical trials (computer vision). • Development of a research article feature extraction tool using NLP models. • Creation of a prompt engineering solution with 150+ categories using the Azure API for OpenAI. 💻: Python, Langchain, OpenCV, Pytorch.

Creation of an educational mobile application on French territory in the market-leading format. • Development of a geographical educational mobile game on French territory with 10K+ downloads. • Collaboration with a Polish company on a C# and Godot project by designing the application from scratch and deploying on PlayStore and AppStore. 💻: C#, Godot, Git.

In the context of insurance fraud detection, development of a model for verifying the accuracy of the bicycle model provided by the user. • Design of a bicycle brand classification pipeline based on user photos. • Data processing and analysis for decision-making, with deep learning on various ML models. 💻: Pytorch, Python, Pandas, Numpy.

Development of a financial dashboard application, creation of the user rights management platform, payment platform with Stripe and data automations. • Collaboration with the CTO to develop a financial dashboard API in Python and FastAPI. • Maintenance and upgrade of the web application in PHP and VueJS. • Creation of test suites and a CI/CD pipeline with GitHub Actions, deployment with AWS. 💻: Python, FastAPI, PHP, VueJS, AWS, Stripe.