$$
{{ $t($store.state.user.experience_value_in_dollars) }}
Junior
{{ $t($store.state.user.experience_search_name) }}
0
jobs
Data scientist / Engineer. Kaggle Competitions Expert. DS teacher @ ISEN Lille & School of AI Paris
Dubreu Benjamin
,
Lille, France
Experience
Other titles
Skills
I'm offering
I create value using data. I am a data scientist who puts himself in his clients' shoes and therefore oriented towards results and careful to provide the best ROI for each and every project.
With a proven record as a Data Scientist (both in industry and international Kaggle competitions), I also bring value with knowledge few standard Data Scientist have: bash, github, docker, airflow. I can therefore build a model, and handle its delivery in production.
With a proven record as a Data Scientist (both in industry and international Kaggle competitions), I also bring value with knowledge few standard Data Scientist have: bash, github, docker, airflow. I can therefore build a model, and handle its delivery in production.
Markets
United Kingdom
Links for more
Once you have created a company account and a job, you can access the profiles links.
Industries
Language
German
Good
English
Fluently
French
Fluently
Ready for
Larger project
Ongoing relation / part-time
Full time contractor
Available
My experience
2020 - ?
job
Deep Learning Teacher
ISEN.
and School of AI (Paris)
Masters degree student body class at ISEN, School of Engineering (Lille) and School of AI
(Paris) Covering Neural Networks from scratch up to CNNs, LSTMs, Autoencoders and GANs
Masters degree student body class at ISEN, School of Engineering (Lille) and School of AI
(Paris) Covering Neural Networks from scratch up to CNNs, LSTMs, Autoencoders and GANs
Deep learning, Engineering, Neural networks, Ai, UP
2018 - ?
job
Data Engineer / Data Scientist
Azure Data Factory.
Fourth client: Total (Energy. France's biggest company and world twentieth)
- Data Engineer: I help process real-time data from drilling platforms all around the world via
Spark Streaming, and trigger an Azure Data Factory when new data is thus collected. This
data is then processed to make several calls to a third-party API, using Azure Functions.
Results of these calls are then stored in Azure Cosmos Databases (NoSQL) for real-time
consumption on the DrillX platform.
Third client: Banque Publique d'Investissement (BPI)
- Data engineer: helped process Data Transfer Objects from frontend to backend of the PGE
(Prêt Garanti par l'Etat) platform to help deliver user's satisfaction metrics. The PGE platform
has helped generate more than 100 billions loans to french companies, and has a NPM of 71.
Environnement: AWS. Language: JAVA.
Second client: Kiabi (fashion/retail)
- Big data engineer: building pipelines using Spark, Hive, Impala on a Cloudera
environnement. There I had the opportunity to discover how cloud environnement are built from scratch, thus gaining a more thorough understanding of the inner workings of cloud
servers.
First client: Adeo Group (3rd company worldwide in the home improvement market):
- Data Science: one year project for prediction of missing/broken items in stores deliveries.
For several Business Units in the group (Leroy Merlin Italy and Russia, BricoMart Spain ).
The model finds 75 % of the missing or broken items, for 43 % of costs of the former
procedure.
Environnement: GCP (Big Query, Virtual Machine, Cloud Functions, Storage)
- POCs: clustering of stores for openings stock prediction, new products sales
forecasting
- Data Engineer: I help process real-time data from drilling platforms all around the world via
Spark Streaming, and trigger an Azure Data Factory when new data is thus collected. This
data is then processed to make several calls to a third-party API, using Azure Functions.
Results of these calls are then stored in Azure Cosmos Databases (NoSQL) for real-time
consumption on the DrillX platform.
Third client: Banque Publique d'Investissement (BPI)
- Data engineer: helped process Data Transfer Objects from frontend to backend of the PGE
(Prêt Garanti par l'Etat) platform to help deliver user's satisfaction metrics. The PGE platform
has helped generate more than 100 billions loans to french companies, and has a NPM of 71.
Environnement: AWS. Language: JAVA.
Second client: Kiabi (fashion/retail)
- Big data engineer: building pipelines using Spark, Hive, Impala on a Cloudera
environnement. There I had the opportunity to discover how cloud environnement are built from scratch, thus gaining a more thorough understanding of the inner workings of cloud
servers.
First client: Adeo Group (3rd company worldwide in the home improvement market):
- Data Science: one year project for prediction of missing/broken items in stores deliveries.
For several Business Units in the group (Leroy Merlin Italy and Russia, BricoMart Spain ).
The model finds 75 % of the missing or broken items, for 43 % of costs of the former
procedure.
Environnement: GCP (Big Query, Virtual Machine, Cloud Functions, Storage)
- POCs: clustering of stores for openings stock prediction, new products sales
forecasting
Sales, Platform, Cloud Functions, Drilling, Backend, Drilling, Energy, Spark Streaming, Energy, Storage, Science, Streaming, Hive, Spark, Java, NoSQL, Cloud, Fashion, Big Data, Azure, Forecasting, AWS, Frontend, Retail, Data Science, Backend, API
2019 - 2020
job
Kaggle Expert
State-of-the Art.
Several Competions, three medals using Deep Learning (Pytorch):
- Natural Language Processing: Jigsaw unintended bias in Toxicity Classification
rank: 127/3165 (top 5%: silver medal); score: 0.943 (top score: 0.947)
Main challenges: using Bert, a then State-of-the Art new Deep Learning architecture
- Computer Vision: RSNA Intracranial Hemorrhage Detection
rank: 84/1345 (top 7%: bronze medal); score: 0.055 (top score: 0.043 (the less the better))
Main challenges: (data engineering) pre-processing the scans into data that is consumable by
a CNN architecture
- Computer Vision: PANDA prostate-cancer-grade-assessment
rank: 58/1030 (top 6%: bronze medal); score: 0.921 (top score: 0.94)
Main challenges: (data engineering) pre-processing biopsies stored in .tiff format, with single
images up to 35000x25000 pixels. That required a pipeline that identifies relevant tissue part
(most of the biopsy is just whitbackground white), split them into square tiles, and pass them to the models as batches of pack of tiles (as opposed to batches of image).
- Natural Language Processing: Jigsaw unintended bias in Toxicity Classification
rank: 127/3165 (top 5%: silver medal); score: 0.943 (top score: 0.947)
Main challenges: using Bert, a then State-of-the Art new Deep Learning architecture
- Computer Vision: RSNA Intracranial Hemorrhage Detection
rank: 84/1345 (top 7%: bronze medal); score: 0.055 (top score: 0.043 (the less the better))
Main challenges: (data engineering) pre-processing the scans into data that is consumable by
a CNN architecture
- Computer Vision: PANDA prostate-cancer-grade-assessment
rank: 58/1030 (top 6%: bronze medal); score: 0.921 (top score: 0.94)
Main challenges: (data engineering) pre-processing biopsies stored in .tiff format, with single
images up to 35000x25000 pixels. That required a pipeline that identifies relevant tissue part
(most of the biopsy is just whitbackground white), split them into square tiles, and pass them to the models as batches of pack of tiles (as opposed to batches of image).
Deep learning, Computer vision, Architecture, Data engineering, Natural, Less, Engineering, Assessment, Processing, UP
2016 - 2017
temp
Political Science PhD Student
Lille University.
Collected data on income distribution and electoral participation of 40 OECD countries.
Used this data to measure the impact of economic inequalities on democratic participation.
Teaching duties: political science, statistics " 101 "; high school intro to economics
Before Various experiences: Public relations agency / elected officials' staffer
I helped organize several runs for office in 2014. Planning an election is a big project, with lots of technicalities (how much can you spend ? Who can give you money ? What kind of add are you allowed to run ?). You build a retroplanning from the election date up to twelve
months before it. It requires the ability to anticipate roadblocks, to talk to a lot of different stakeholders with various interests, and to recruit and organize volunteers. I will be
able to use all those skills as a Project Manager.
KAGGLE
Kaggle competitions are 4 months data science and engineering projects where individuals help
solve a state of the art analytics problem. As there are always more ideas to try than time to test
them, competitors must prioritize their work. Most of the time the difference is made not by
models, but by mastery of Python for the preprocessing parts.
Used this data to measure the impact of economic inequalities on democratic participation.
Teaching duties: political science, statistics " 101 "; high school intro to economics
Before Various experiences: Public relations agency / elected officials' staffer
I helped organize several runs for office in 2014. Planning an election is a big project, with lots of technicalities (how much can you spend ? Who can give you money ? What kind of add are you allowed to run ?). You build a retroplanning from the election date up to twelve
months before it. It requires the ability to anticipate roadblocks, to talk to a lot of different stakeholders with various interests, and to recruit and organize volunteers. I will be
able to use all those skills as a Project Manager.
KAGGLE
Kaggle competitions are 4 months data science and engineering projects where individuals help
solve a state of the art analytics problem. As there are always more ideas to try than time to test
them, competitors must prioritize their work. Most of the time the difference is made not by
models, but by mastery of Python for the preprocessing parts.
Python, Project Manager, Public relations, Data Science, Teaching, Analytics, Test, Engineering, Statistics, It, Science, Office, UP, Manager
My education
2017
-
2019
Supélec
Masters, Machine Learning Engineer
Masters, Machine Learning Engineer
2017
-
2019
High School Spanish
N/a, English
N/a, English
University of California San Diego
N/a, Maths (Minor in Computer Science)
N/a, Maths (Minor in Computer Science)
?
-
2015
Minot State
N/a, N/a
N/a, N/a
?
-
2013
Lille University
Masters, Political Sciences
Masters, Political Sciences
Dubreu's reviews
Dubreu has not received any reviews on Worksome.
Contact Dubreu Benjamin
Worksome removes the expensive intermediaries and gives you direct contact with relevant talent.
Create a login and get the opportunity to write to Dubreu directly in Worksome.
are ready to help you
and get specific bids from skilled talent in Denmark