Co-built a web app for exploring French agricultural parcel data through a conversational chatbot. Users type natural language prompts to query communes, crop types, years and spatial layouts, the chatbot converts them to SQL on the fly and automatically displays the results as maps, graphs and tables. Also supports result comparison across years and data upload/download. Built with Python, SQ, GeoPandas and Huggingface.

Analyzed water quality data to answer questions about water quality and spatial patterns: engineered temporal features, ran K-Means, validated clusters with I2M2, and mapped spatial patterns with CORINE land cover.

Built 5-year equity datasets, trained global and per-stock deep learning models, and benchmarked architectures to highlight the Transformer as the most accurate.

Estimated battery State of Health from MIT telemetry, comparing recurrent networks, ensemble methods, and regression models for accuracy and interpretability.

Labeled agricultural plots by fusing monthly satellite imagery, evaluating multiple deep learning classifiers and ensembles for seasonal robustness.

Implemented a Flask UI to analyse circular codes, generate directed graphs, and surface strong comma-free code families for bioinformatics research.

Created a Node.js platform for generating short links, storing usage in MongoDB, and monitoring redirection metrics via a custom dashboard.

Rebuilt the Konkr.io strategy game in C++, including procedural map generation, full combat rules, and AI opponents.