Bienvenue sur le profil Malt de Sofian !
Localisation et déplacement
- Localisation
- Paris, France
- Télétravail
- Effectue ses missions majoritairement à distance
Vérifications
Charte du freelance Malt signée
Consulter la charte
Influence
Chargement...
Langues
Catégories
Compétences (5)
- Data Science
-
- Tous
-
Sofian en quelques mots
Ingénieur diplômé de l'ENSTA Paris et du Master MVA de l'ENS Paris-Saclay,
Je suis formé et experimenté en Machine Learning, Deep Learning, NLP et vision.
Je suis actuellement ingénieur de recherche au Noah's Ark Lab où je travaille sur des projets de recherche en deep learning, RL, bandits et optimisation.
Je suis formé et experimenté en Machine Learning, Deep Learning, NLP et vision.
Je suis actuellement ingénieur de recherche au Noah's Ark Lab où je travaille sur des projets de recherche en deep learning, RL, bandits et optimisation.
Expériences
Huawei
Télécommunications
Research Engineer - En tant que freelance
Noah's Ark Lab
Crédit Agricole SA - Groupe Crédit Agricole
Banque & assurances
Data Scientist - En tant que freelance
DataLab, Team Semantica.
- Aspect Based Sentiment Analysis for clients' feedback analysis :
The goal of this project was to improve the model in production by using transformers models.
A multitask model that achieves both aspect detection and polarity detection was implemented and put in production.
- Project on financial contracts analysis from the investment bank CACIB :
The goal of the project is to build a search engine on the contracts database.
The contracts are in pdf format, have to be converted to images, ocerized and indexed in an Elasticsearch index.
The search engine has many features : - Segmentation of the contract into paragraphs, lists, clauses
- Clause classification that allows to extract specific clauses of interest using tf-idf approaches.
- Extraction of relevant spans that helps rule whether the contract is transferable using techniques inspired from neural question answering and semantic similarity
- Training of Multimodal Classification deep learning models of car insurance contracts using textual (ocr) and visual content.
- Training of Information Extraction deep learning models from car insurance contracts using semantic segmentation and textual embeddings.
Technological stack : pytorch, tensoflow, transformers, tesseract, elasticsearch, mlflow, gitlab CI/CD, poetry, docker, AWS S3, mlflow, etc.
- Aspect Based Sentiment Analysis for clients' feedback analysis :
The goal of this project was to improve the model in production by using transformers models.
A multitask model that achieves both aspect detection and polarity detection was implemented and put in production.
- Project on financial contracts analysis from the investment bank CACIB :
The goal of the project is to build a search engine on the contracts database.
The contracts are in pdf format, have to be converted to images, ocerized and indexed in an Elasticsearch index.
The search engine has many features : - Segmentation of the contract into paragraphs, lists, clauses
- Clause classification that allows to extract specific clauses of interest using tf-idf approaches.
- Extraction of relevant spans that helps rule whether the contract is transferable using techniques inspired from neural question answering and semantic similarity
- Training of Multimodal Classification deep learning models of car insurance contracts using textual (ocr) and visual content.
- Training of Information Extraction deep learning models from car insurance contracts using semantic segmentation and textual embeddings.
Technological stack : pytorch, tensoflow, transformers, tesseract, elasticsearch, mlflow, gitlab CI/CD, poetry, docker, AWS S3, mlflow, etc.
Crédit Agricole SA - Groupe Crédit Agricole
Banque & assurances
NLP research intern
DataLab, Team Semantica.
Designed a French textual search engine with a span extraction module.
Achieved state-of-the-art results on the Phrase-Indexed QA (PIQA) benchmark.
Achieved state-of-the-art results on the squad-open benchmark.
(preprint : https://arxiv.org/abs/2012.09766)
Designed a French textual search engine with a span extraction module.
Achieved state-of-the-art results on the Phrase-Indexed QA (PIQA) benchmark.
Achieved state-of-the-art results on the squad-open benchmark.
(preprint : https://arxiv.org/abs/2012.09766)
Recommandations externes
Consultez les recommandations qu'a reçues Sofian