Freelancer profile translated to English.

Description

Data scientist & AI with two years of experience in R&D at STMicroelectronics as a Data Scientist/MLops Engineer, specializing in the development of artificial intelligence and NLP solutions based on LLMs, CNNs, and ViTs, as well as continuous integration of machine learning models. Expertise in code optimization citing quantization, pruning, and knowledge distillation, and the implementation of LLM and ViT models. Demonstrated ability to solve complex problems related to firmware integration on resource-constrained devices, with a focus on energy efficiency and performance. Shows exceptional creativity in developing innovative solutions and a strong aptitude for research.

Languages

French
Native or bilingual
English
Native or bilingual
Arabic
Native or bilingual

Workplace preferences

Can work on-site

Paris (up to 50km)

STMicroelectronics
Data Scientist & Python/AI Dev
TECH
January 2023 - September 2023 (8 months)
Paris, France
Missions
• Development of a solution based on a Time-of-Flight (ToF) sensor for Deep Neural Network (DNN) applications.
• Optimization of deep learning models for deployment on STM32 microcontrollers.
• Evaluation and improvement of model accuracy after optimization.
• Investigation of new optimization techniques to further reduce energy consumption without compromising performance. Key achievements:
• Reduction of model size and energy consumption through quantization and the use of the CubeAI extension.
• Improvement of optimized model accuracy through advanced Fine-Tuning methods.
• Visualization of outputs and predictions on a dashboard connected to the MCU via WI-FI
STMicroelectronics
AI & MLOPS Engineer R&D
TECH
September 2023 - May 2025 (1 year and 8 months)
Paris, France
• Design of a hybrid Edge-Cloud (AWS) MLOps pipeline for the deployment of Deep Learning models (CNN, Transformers) with optimization of data transfers and cloud cost management.
• Quantization and pruning of models for embedded targets with management of memory constraints (<2MB RAM) and inference optimization.
• Inference acceleration via CMSIS-NN (latency reduction to 15ms).
• Implementation of CI/CD automation and model versioning with MLflow. Key achievements:
• Design and adaptation of ST ModelZoo models on the "Hand posture recognition" application and the application of neural architecture search (NAS).
• Real-time dashboard with Grafana and Prometheus.
• Automation of Integrations and deployments via AWS CloudFormation and Terraform
• Average latency reduced to 15ms on FreeRTOS target with CMSIS acceleration.