Praktika in Deutschland

Très bonne entreprise, très compréhensive et laisse beaucoup de libertés pour décider comment on s'organise.

Point négatif, peu d'accompagnement (mais Onboarding tout de même et ouverts à aider si on demandait de l'aide)

• Accelerated ML model inference by (OpenAI CLIP 13x, Florence 4x) through CUDA-enabled Docker containers
running on cost effective AWS GPU instances.
• Improved batching capabilities from one image per batch on GPU to sixty-four by building GPU
memory/performance monitoring tools and implementing Pytorch Dataloaders.
• Constructed ETL pipelines to extract OpenSearch embeddings. Enhanced runtime of dimensionality reduction
and clustering algorithms with RAPIDS cuML GPU libraries. Helped create POC for interactive Voxel51
visualizations.
• Managed AWS PostgreSQL databases using SQL Alchemy and Pandas. Upgraded Grafana and Dash dashboards
with real-time metrics. Developed in-house reverse geocoding algorithm, accelerated with multiprocessing,
helped verify data validity.