Logo Charly Hayoz

Model training - From architecture to performance

Training is at the heart of every AI project. I design models adapted to your data and objectives, combining scientific rigor and mastery of modern tools.

  • Choice of architecture

Selection according to the nature of the problem: CNN, RNN, LSTM, Transformer, diffusion models.

  • Data pre-processing

Standardization, augmentation, split train/test to guarantee data quality and representativeness.

  • Model optimization

Use of AdamW, learning rate management, gradient clipping to stabilize and accelerate learning.

  • Distributed training

Scalability with multi-GPU, TPU, DeepSpeed, FSDP for large or complex models.

  • Experience follow-up

Tracking of runs, hyperparameters and metrics via Weights & Biases, MLflow for optimum reproducibility.

My expertise

AI development - Custom architecture, integration and performance

AI Development

Creation of intelligent architectures, APIs, conversational agents, recommendation systems

Language Models (LLM) - Automate, understand and generate with precision

LLM & NLP

Integration of models such as GPT, LLaMA, Mistral, Claude, etc. into business workflows

Fine-tuning - Adapting AI models to your business data

Fine-tuning & Training

Adaptation of pre-trained models to specific corpora, supervised or reinforcement training

Fine-tuning - Adapting AI models to your business data

Neural networks & machine learning

Design and training of deep learning models (CNN, RNN, Transformers) & machine learning (Random Forest, Scikit-Learn) for complex cases

RAG - Generation augmented by documentary research

Retrieval-Augmented Generation (RAG)

Combining documentary research and generation for precise, contextualized answers

Tailor-made Edge AI - real-time AI optimized for you

Edge AI

Design and deployment of modular, secure local AI architectures capable of processing data directly on the device

MLOps & Deployment - Industrialize and secure your AI models

Deployment & MLOps

Containerization, CI/CD, monitoring, scalability, model security in production