I designed and built a secure Python application that ingests multi-source satellite imagery and uses computer vision plus AI to detect and track signs of military troop movement at global scale. The system supports large GeoTIFF inputs and regional Areas of Interest, enabling analysts to focus on specific theaters while maintaining worldwide coverage.

The pipeline performs robust pre-processing : tiling, denoising, cloud masking, and orthorectification; before running multiple inference stages. Images are normalized and aligned for consistent comparisons across time, sensors, and resolutions, ensuring high-confidence detections in challenging conditions.

For analytics, the app combines object detection (vehicles, encampments, logistics assets), change detection (appearance/disappearance, expansion), and temporal tracking to link observations into movement "tracks." It estimates direction, speed, and confidence, correlating repeated sightings to surface likely routes and corridors.

Results are written to a geospatial database (PostGIS) with fast spatial queries and aggregation for heatmaps, AOI boundary crossings, and route clustering. A secured FastAPI layer exposes endpoints for internal dashboards and automated alerting (e.g., notify when ≥N tracked vehicles cross a boundary within 24 hours).

The system scales via containerized workers for batch and near-real-time processing, with Celery task orchestration and Redis queues. It includes audit logging, reproducible runs, and strict data-handling and access controls suitable for sensitive environments. Confidential work, no public demo or code link.

Key features included :

• Automated ingestion of multi-source satellite imagery (large TIFF/GeoTIFF, multi-band) with AOI filtering and metadata normalization.
• Robust pre-processing: tiling, denoising, cloud masking, orthorectification, and sensor-aware normalization for consistent comparisons.
• Object detection (vehicles, encampments, logistics assets) using deep learning; change detection to highlight new or removed infrastructure.
• Temporal tracking that links detections across time to form movement tracks with estimated direction, speed, and confidence.
• Route and corridor analysis with clustering and heatmaps to reveal habitual paths, chokepoints, and boundary crossings.
• Geospatial indexing (PostGIS) for fast spatial queries, AOI overlays, proximity searches, and aggregation.
• Alerting rules and thresholds (e.g., “notify if ≥N tracked vehicles cross boundary within 24h”), delivered via secure internal endpoints.
• Secure API (FastAPI) powering internal dashboards and tools with role-based access and auditable request logs.
• Scalable processing with containerized workers, Celery orchestration, and Redis queues for batch and near-real-time workloads.
• Compliance and auditability: full run reproducibility, immutable logs, and strict data-handling & access controls for sensitive environments.

I developed QueerView.net, a community-powered web platform where LGBTQIA+ individuals can anonymously review their employers based on inclusivity, respect, safety, and workplace culture. The platform is designed to promote transparency in the workplace and empower marginalized voices through collective experiences.

In addition to the public-facing review system, I also built a dedicated employer portal, allowing organizations to manage their profile, respond to reviews, and demonstrate their commitment to diversity and inclusion. A job board and a blog section round out the platform, offering resources and opportunities tailored to LGBTQIA+ professionals.

Key features included :

• An anonymous review system with moderation tools
• Company profiles with aggregated scores across key diversity and inclusion metrics
• A separate employer portal allowing organizations to respond to reviews and showcase their commitment to LGBTQIA+ inclusivity
• A responsive, accessible UI optimized for both desktop and mobile users

Coronavir.org was a rapidly developed public-facing information hub created at the onset of the COVID-19 pandemic. Built in under a week, the platform offered live updates of global and country-specific case numbers, deaths, and recoveries. It automatically parsed and stored official data daily, generating dynamic charts and key health metrics. Custom calculations and predictive models helped provide clearer context and early projections for visitors seeking accurate, up-to-date information.

Beyond raw data, the site aggregated and categorized scientific publications related to the virus to make them easily accessible to the general public. A built-in CMS allowed the client to create and manage informational pages independently, while a moderated forum gave users a space to share updates, ask questions, and engage in discussion.

Key features:

• Automated parsing and visualization of COVID-19 statistics
• Predictive trend calculations and real-time charts
• Scientific article aggregation and structured display
• CMS for custom content management and editing
• Public forum for community discussion and updates

Gearscore was a complex data-driven platform designed to unify the ecosystem of outdoor sports gear, from manufacturers to reviewers to marketplaces. The platform focused first on skis and snowboards, building a rich database by parsing and matching product data across Google Shopping, marketplace APIs, brand websites, and review sources. Each product entry was enriched with GTINs, SKUs, pricing, availability, and expert reviews. Users could visit a public-facing website to explore detailed product listings, compare offers, read aggregated reviews, and discover where to buy the gear at the best price.

The project also included a monetization layer: a lightweight JavaScript widget for reviewers to embed into their articles. This widget dynamically displayed live deals and prices for the reviewed products, driving traffic to partner marketplaces. A tracking system recorded impressions, clicks, and purchases, allowing Gearscore to handle payouts to reviewers and invoicing for marketplace partners. Each brand, reviewer, and merchant had access to a personalized dashboard with campaign metrics, engagement analytics, and revenue breakdowns.

Key features:

• Product data aggregation from marketplaces, brands, and shopping feeds
• Central product matching engine based on GTINs and SKUs
• Aggregated reviews from online and offline sources
• Reviewer widget with tracking and affiliate monetization
• Personalized dashboards for brands, marketplaces, and reviewers
• Full admin dashboard with advanced metrics and campaign controls

I contributed to the development and maintenance of MediaVacances.com, a long-standing vacation rental website offering direct bookings between property owners and travelers across Europe.

My work involved improving both the back-end architecture and user-facing features of this large-scale web platform. Key tasks included :

• Refactoring legacy code for better performance, maintainability, and scalability
• Enhancing search and filtering functionality for listings
• Implementing updates to the booking and messaging systems
• Optimizing page speed and responsiveness across devices

As part of the broader Muteco initiative led by MEDEF, I developed a web application designed to assist surveyors during on-site visits to companies. The tool was used on tablets to collect structured data through a series of dynamic forms and questionnaires.

Once the forms were completed, the app automatically generated custom reports with calculated indicators, summary tables, and data visualizations to help identify workforce trends, training needs, and economic challenges at the company level.

Key aspects of the app included: :

• Touch-optimized mobile-friendly interface
• Complex form workflows with validation and conditional logic
• Real-time data processing and automated chart generation
• Offline support and reliable sync mechanisms

I developed a custom web application for IRFO (Institut des Rencontres et de la Forme) to support a company-wide pedometer challenge aimed at encouraging physical activity among employees and partner organizations.

The platform automatically synced daily pedometer data and featured a robust admin dashboard for managing custom challenges, both global and per-company. Users could track performance through leaderboards, team rankings, and interactive graphs displaying metrics such as steps walked, calories burned, and overall engagement.

Built with a focus on performance and usability, the app included :

• Real-time data visualization
• Custom challenge logic and filtering
• Secure authentication and role-based access
• Clean, responsive UI for desktop and mobile users

My first client ! A custom web platform for Rivages Propres, a French environmental NGO dedicated to combating coastal pollution through organized beach cleanups. The solution streamlined their volunteer coordination, event management, and data tracking, replacing manual processes with an intuitive, CMS-like system. It empowered administrators to create cleanup campaigns, manage registrations, and analyze waste collection metrics, all while ensuring accessibility for non-technical users.

The platform featured a dynamic map interface to visualize cleanup sites and collected waste data, reinforcing the NGO's mission with real-time impact metrics. I implemented role-based dashboards for organizers, volunteers, and partners, alongside tools to generate PDF reports and public-facing transparency dashboards. Emphasis on mobile responsiveness allowed volunteers to engage seamlessly from any device during fieldwork.

Key aspects of the project included:

• Custom CMS for content updates and event management without coding
• Geolocation integration for mapping cleanup zones and volunteer check-ins
• Data visualization (charts/graphs) to showcase environmental impact
• Secure authentication for organizers and volunteer accounts