Tech stack:
Python
Python
Docker
Docker
FastAPI
FastAPI
Tensorflow
Tensorflow
Story
While researching ideas for business end-user components of Fullstack MLOps Platform for Formula 1, I stumbled upon intriguing community materials:
- a complete dataset of F1 cars images, created by José Henrique Brito and Sérgio Gomes,
- a PoC code repository and an article published on Kaggle, both submitted by Faw.
In my opinion: these looked promising… and could be combined, in order to create a production-scale solution.
Solution
Having the above in mind, the scope of work turned out to be quite extensive – the PoC code needed:
-
- tidying up and integrating dataset into existing Faw’s PoC code repository,
- training the Tensorflow/Keras model and saving it,
- connecting above runs to MLFlow instance backed by MinIO S3 object storage,
- adding API endpoints with FastAPI framework,
- working out data visualization principles for created predictions,
- designing website layout with Jinja templates,
- splitting code into Python modules,
- adding proper documentation of the whole solution.
Screenshots
Lesson Learned
Improving my skills was an obvious award. But what made me really proud of this activity were Faw’s own words in one of the pull requests: “(…) learning so much from your commits and code, just wanted to let you know it is very appreciated!”. That’s why I’m doing it – and giving back to the community.
Statistics
-
- 2: developers,
- 9: tags,
- 32: days (only 2.5 Scrum sprints!),
- 51: commits,
- 2068: lines of code, out of which
- 1663: were removed,
- 3788: were added.