Romaric Sokoundjou Takam

Geographic data processing and standardization for last-mile logistics (2 countries, 185,000 users, 5To geographical data processed) • Design and deployment of a multi-source georeferencing engine (World Bank, Google Maps, OpenStreetMap, IGN) to generate digital addresses compliant with the S42 standard. • Analysis of geographic and business data related to the company's activity. • Recruitment and management of a hybrid team of 5 developers using agile methodology (SCRUM). • Development of a deep learning model for inferring geomatic data from Sentinel satellite images • Development of a WhatsApp chatbot for customer service (reducing the average ticket processing time by 29.3%) Technical environment: Python, PostGIS, PgVector, Pydantic AI, LangChain, OpenStreetMap, Django, SQLAlchemy/SqlModel, ReactJS, Htmx/AlpineJS, FastAPI, Google Cloud (Maps API, Storage, Run, Task, Scheduler), Firebase, PyTorch, PostgreSQL 17, Docker, Streamlit, PostHog, Pydantic LogFire

Online CTR Prediction, Topic Modeling, Brand Safety, Anomaly Detection, Optimal Control, Media Attribution, MLOps. (4B + events per day)

Realtime churn prediction machine-learning algorithms on all SFR mobile clients (~15 million users)

- Newspaper articles recommender system - Marketing Mix Modeling (experiments and media attribution for multitouchpoint ads by different ad partners, online and offline)

Image classification algorithms to help horse riders train by realtime tracking, comparing their sessions against the "best" ones and suggesting training exercices

General Game Playing & UCT Monte-Carlo algorithms for the Go Game