Freelancer profile translated to English.
Back to original languageAbout Hugo
🧠 Artificial Intelligence Engineer with 4+ years of experience, I design tailor-made solutions in NLP, LLM, ASR, and RAG to automate business processes, optimize performance, and create growth levers through data.
🚀 After a background combining applied research and industry (Amazon Alexa AI), I founded Yula Studio, where I help companies integrate the latest advances in generative AI (GPT, LLaMA, Mistral, AI agents, vector databases) into their existing tools and systems.
🔍 Key Expertise:
• NLP / LLM: email classification, information extraction, text generation, fine-tuning of pre-trained models (GPT, BERT, LLaMA, Mistral)
• ASR / Speech-to-Text: low-resource models, multilingual adaptation (Alexa Arabic, FireTV), rescoring, data augmentation
• RAG & AI Agents: vector search, data-augmented generation, intelligent multi-agent systems (LangChain, Pinecone)
• ML & Data Engineering: processing pipelines, database structuring, supervised and unsupervised models
🛠️ Technical Stack:
Python (PyTorch, TensorFlow, LangChain, FastAPI) – AWS (SageMaker, EC2) – Docker – Pinecone – Neo4j – React – Heroku
🤝 I am involved in the entire lifecycle of an AI project: scoping, prototyping, production deployment, and post-delivery support. Comfortable working independently or integrated into your teams, I adapt to your business and technical challenges.
📬 Let's discuss your AI needs: a quick audit, a prototype, or a turnkey integration? I offer you a pragmatic and personalized solution.
French
Native or bilingual
English
Native or bilingual
Spanish
Native or bilingual
Can work on-site
Paris (up to 50km)
Experience
- YulaChief Technology OfficerNovember 2023 - Today (2 years and 7 months)Paris, FranceDevelopment of RAG (Retrieval-Augmented Generation) systems to automate e-commerce customer service with LLMs• Design and deployment of a RAG (Retrieval-Augmented Generation) solution based on large language models (LLMs) to automate responses to customer inquiries, with a measurable impact on operational efficiency, response accuracy, and user satisfaction.• Creation of a complete NLP data pipeline for extraction, cleaning, indexing, and vector storage of textual content, ensuring fast and relevant document retrieval.• Implementation of a modular architecture combining multi-agent systems and LLMs for intelligent management of complex user interactions.• Recruitment and coordination of a multidisciplinary team (Front-End, Full Stack, Machine Learning Engineer) to ensure the design, deployment, and continuous improvement of AI services in production.• Technical guidance on architectural choices and collaborative development best practices for applications combining artificial intelligence and high-performance web interfaces.Technologies: LangChain, Pinecone, FastAPI, React, Remix, HerokuKeywords: RAG, LLM, NLP, retrieval-augmented generation, multi-agent systems, chatbot, vector database, Python, FastAPI, LangChain, React, Heroku, e-commerce automation
- AmazonApplied ScientistAugust 2021 - November 2023 (2 years and 3 months)London, UKExpert in Automatic Speech Recognition (ASR) specializing in low-resource environments• Development of high-performance RNN-T ASR models in low-resource contexts, integrating heterogeneous data (real and synthetic data), resulting in an 83% relative Word Error Rate (WER) reduction.• Implementation of an incremental learning framework reducing training costs by 75%, optimizing production iterations.• Specialization in domain and localization adaptation through synthetic audio data generation, for voice search systems based on pre-trained models.• Performance optimization using advanced rescoring techniques: first pass shallow fusion, Internal Language Model Estimation (ILME), RescoreBERT, achieving an 8% relative WER reduction.• Proficiency in data preparation, audio data augmentation, transfer learning, incremental learning, and fine-tuning of ASR models. Experience in data source weight mixing to improve model robustness and accuracy.Technologies: RNN-T, Whisper, Kaldi, ESPnet, HuggingFace Transformers, PyTorch, TensorFlow, wav2vec 2.0, BERT, RescoreBERTKeywords: speech recognition, ASR, WER, RNN-T, domain adaptation, language model fusion, synthetic audio, low-resource ASR, incremental learning, fine-tuning, audio data augmentation
- AmazonApplied ScientistFebruary 2021 - July 2021 (5 months)Barcelona, SpainApplied research on the scalability of MARL (Multi-Agent Reinforcement Learning) models with a dynamic number of agents• Advanced study on the scalability of multi-agent reinforcement learning (MARL) models in contexts where the number of agents varies at inference time, a key challenge for real-time applications (logistics, traffic, robotics).• Implementation of state-of-the-art collaborative algorithms such as CommNet and BiCNet, as well as several reference baselines, to evaluate their robustness and efficiency in complex multi-agent environments.• Design of a realistic testbed based on an autonomous traffic simulation, allowing for performance comparison in terms of coordination, scalability, and convergence time.• Presentation of results and analyses at an internal conference, highlighting implications for applied research in distributed collective intelligence.Technologies: Ray RLlib, OpenAI Gym, PyTorch, TensorFlow, Docker, AWS (SageMaker, EC2)Keywords: MARL, reinforcement learning, multi-agent systems, CommNet, BiCNet, collaborative AI, traffic simulation, RLlib, Gym, AWS SageMaker, Docker, PyTorch, TensorFlow
Recommendations
Be the first to recommend Hugo
Help this freelancer shine by sharing your experience working together.
These freelancer profiles also match your criteria
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4
Education
- Software EngineeringTianjin University2021Software Engineering
- Master of ScienceESILV - Engineering School Leonardo da Vinci2021Master's degree, Data Sciences