Paul Chauvin

Working on GenAi models applied to chemistry and biology. Building and finetuning LLMs. Creating AI products at scale to help our chemists be more efficient in their work.

• Rakam Ai - mission for Evalucar: Built a classifier for detecting any car from an image or screenshot (brand, model, generation, body type). Developed the entire pipeline: data processing, fine-tuning of LLMs (Mixtral 8x7B, Mistral 7B, Arctic), created a vision model for classifying car elements (implemented ResNet, ViT, and VGG architectures), compared model performances, and deployed in production using Docker, Poetry, PyTorch, and AWS. • Campsider: Developed an app giving accurate pricing of second-hand products. Gathered data by developing web scrapers leveraging images and text, used LLMs for data formatting, and created comparison tools to refine product pricing. Sold the product as a SaaS model. • Believe Corporate: Provided expertise in a call center to evaluate the potential integration of AI and GenAI tools. Proposed an action plan for the integration of these projects and estimated the associated financial gains. • Ako: Assisted in the initial technical setup for sales prediction in the retail sector. Developed an ARIMA model. • Musically: Built a voice chatbot specialized on music.

Part of data science team. Developed transformer-based models for classifying traffic connections and improving NanoCorp's cyberthreat detection model. Created end-to-end ML pipeline (web data scrapper, DBT, ClickHouse, Docker, MLflow)

Deezer is one of the top leading music-streaming platforms in the world. Part of R&D team. Mission working on Deezer's Lyrics-alignment algorithm: • Deep-diving into the existing supervised algorithm in order to understand when the model fails • Literature review on existing unsupervised alignment technics • Developing unsupervised metric for lyrics-alignment model evaluation

Stuart is a European on-demand delivery platform for businesses that aims to revolutionize urban logistics by providing fast, efficient, and eco-friendly delivery services while prioritizing innovation, simplicity, flexibility, and a customer-centric approach. Permanent job: in charge of data models implementation for Stuart's expansion throughout France. • Working on classification Machine Learning models to improve Estimated Time of Arrival for deliveries (improving by 40% the accuracy on targeted deliveries’ clusters) • Dynamic pricing for drivers’ pay depending on special situation (weather, events, month or weekday, supply/demand equilibrium) • Forecasting model for the number of deliveries in every French city • Participating in the migration of all scripts and dashboards used internally due to a major data warehouse change

Better World is a technology company that provides a comprehensive review analysis solution from data collection to visualization, leveraging their proprietary award-winning AI/NLP algorithms. 6-month end-of-engineering-studies internship, leading several missions: • NLP mission in Python: detecting opposed-sentiment verbatims with Scikit-Learn, NLTK libraries, and Bag Of Words (reaching 85% of accuracy, 87% recall for automatic detection) • Data operation missions: Improving the speed of operational processes. Head of projects for several clients. Development of new industries covered by Better World, building ontologies, and improving Machine Learning classification algorithms using MongoDB. • Business analyst missions: determining key sectors to develop Better World's business.

• Coaching and teaching students 10 hours/week • Teaching Computer Science, Data Science, Mathematics, Physics to high school and undergraduate students • Developing efficient way of improving scientific skills to 60+ students

FLYING WHALES is a French company established in 2012 to develop and industrialize a 60 tons Large Capacity Airship. • Part of the propulsion team, leading aeroacoustics project on LCA60T airship • Research work on the aeroacoustic impact of a large number of engines on an airship • Calculation and modelization of airship’s 32 propellers’ aeroacoustic using Matlab

Worker on 4 different departments of the shipyard : • Carbon lamination department • Deck department • Carpentry department • System department