• Implement the ML life cycle management infrastructure for Computer Vision AI models in a distributed architecture supporting multiple edge GPU devices, Migrate inference directly on the sensor, improve latency and quality of frames, reduce network load and GCP bill

• Automate the deployment process and scalability on demand of production environments in GCP and the Custom Nvidia Jetson devices using Ansible, Terraform, Docker and Balena

• Provide technical customer support on API usage and integration, optimize existing code

• Add metadata extraction features for speed, duration, direction, CO2 emissions for traffic analysis, and design with RD the MVP architecture for sensor combining Lidar/GPU/Video

Stacks: CI/CD, GCP, AWS, Terraform, Ansible, Airflow, MLflow, DVC, Nvidia Docker, Cuda GPU, Lidar

• Coordinate and develop the MonkApp MVP, their legacy mobile application allowing AI-assisted vehicle inspection, taking guided photos, displaying damages detected by the API in real time, allowing users to view, sign and export PDF analysis reports damage

• Design with a new team an open-source SDK for UX/UI MonkJS, code the functional aspect, backend integration, annotation modules and 360 visualization of damage

• Iterate with data scientists and the delivery team on the feature roadmap and integrate urgent features required by the front end, e.g. OCR, PDF generation

• Migrate API code and data from Firebase Cloud functions (NodeJs) and Firestore to the new API, develop a dashboard named Dataguru for Image/Train datasets managements

Stack: React, Mobile, TypeScript, NodeJs, Expo, Jira, PDF, OCR, Python, Flask, Postgres, CircleCI Python, Flask, REST, PostgreSQL, SQLAlchemy, Docker, CircleCI, GCP, Auth0, Git

Led the development of the core video processing product for real-time traffic analysis using Computer Vision. This involved

• In Backend: Code and architect vision model pipelines, transport of frames and metadata, concurrently use GPU to process frames from multiples video streams devices on GCP

• In DevOps: Set up the CI/CD stack for the deployment of Docker images on GCP, AWS and optimize the microservices cycle based on resource usage

• In Frontend: Code the Viize Camvas dashboard to display video streams, history and analysis results in real time, manage sensors and annotate regions of interest (ROI)

Stack: Python, FastAPI, PyTorch, ZeroMQ, FFMPEG, OpenCV, WebSocket, Images processing, NginX, Docker, Circle CI, Ansible, React, Next JS, Canvas, WebGL, Pixi JS, TailwindCss, NextJs