Smart me up - Marreli

Private dashboard for autopiloted car fleet

Client

Smart me up - Marreli

Work type

Frontend & Backend

Dates

July 2020 - September 2020

Technical stack

Vue.JS

Python

MongoDB

Typescript

Jango

three.js

SmartMeUp, founded in 2012 and based in Meylan, France, specialized in developing perception software for autonomous driving, focusing on low-power object detection and 3D environment modeling. With my help we created an admin dashboard and reviewed their anotation plateform and system

Challenge

For a short-term mission, I worked independently at Smart Me Up to develop a completely new dashboard for their autonomous car fleet. The primary need was to provide a global overview of the fleet within minutes while also allowing detailed information retrieval for each individual car. In addition to the dashboard, I revamped their data annotation system using Three.js to make it more efficient and user-friendly for external annotators. This required significant improvements to the interface and functionality, streamlining the annotation process. While the new dashboard was straightforward to develop, reworking the annotation system posed a greater challenge, as it required in-depth knowledge of geometry and Three.js. I quickly acquired this expertise on the fly, successfully delivering a robust and enhanced solution.

Work

As a web developer, my primary mission was to design and develop a completely new internal dashboard for managing an autonomous car fleet. The dashboard included features such as real-time geolocation, data aggregation, error reporting, detailed car views, and more. Since the CTO had limited time, I independently led the project, conceptualizing and implementing all features without detailed requirements. This required me to deeply understand Smart Me Up's needs, particularly from a machine learning scientist's perspective, to determine the types of data they required. The dashboard was developed using Django (Python) for the backend, Vue.js (v2) for the front end, and MongoDB for the database. With complete freedom over the product, I handled all aspects of the project, from design to deployment, including managing the release pipeline. This autonomy pushed me to learn and perform tasks beyond traditional development roles. The second part of my role involved revamping the company's data annotation system, used by external freelancers. This system was designed to simplify the process of annotating images from the car fleet, which were later reviewed by the Smart Me Up admin team and integrated into their machine learning models upon validation. Since annotations had to be translated from 2D to 3D, I built the system using Three.js. The annotations included critical data such as identifying cars, humans, traffic signs, and other objects essential for driving. The PhD team at Smart Me Up provided formulas and data formats, which I integrated into the system. Once images were annotated, they were saved to the server for later review and validation by the team.

Result and impact

  • Designed and developed a comprehensive dashboard to manage the autonomous car fleet.

  • Key features included real-time geolocation, data aggregation, error reporting, car detail views, and more.

  • Successfully led the project independently, from concept to deployment, addressing the company's needs without detailed specifications.

  • Utilized Django (Python) for the backend, Vue.js (v2) for the front end, and MongoDB for data storage.

  • Managed the full product lifecycle, including design, development, and release pipeline setup.

  • Revamped the annotation system for external freelancers, enhancing usability and efficiency.

  • Built the system with Three.js to translate annotations from 2D to 3D, incorporating complex geometry calculations.

  • Enabled annotators to classify and label objects such as cars, humans, and traffic signs, critical for autonomous driving.

  • Integrated formulas and data formats provided by the research team, ensuring alignment with machine learning model requirements.

  • Streamlined the process for saving and validating annotated images, facilitating smooth integration into machine learning workflows.

Have a Project ?

Let's discuss together!

malt profile pagelinkedin profile pagegithub profile page