Pierre-Alexandre Broux

Situation: Entrepreneurial initiative to launch a SaaS solution aimed at reducing the time spent in traditional meetings. Task: Design an asynchronous meeting platform incorporating best practices to optimize time management. Actions: • Complete solution development with Django (Python), Htmx and Tailwind CSS. • Deployment and management of infrastructure on Amazon Web Service (AWS) and DigitalOcean. • PostgreSQL database management. • Application of DevOps best practices such as CI/CD and testing (unit, integration and system tests). • Documenting algorithm designs to facilitate knowledge sharing and ensure reproducibility. Result: Significant reduction in the number and duration of meetings, with a significant improvement in user productivity. Stack : Python, Django, Amazon Web Service, Docker, Pytest, Tailwind CSS, Htmx, DigitalOcean, Git, Github Actions CI/CD, PostgreSQL, JavaScript.

Situation: The company wanted to create an artificial intelligence (AI) laboratory dedicated to the development of specialized solutions for recognizing emotions from text (NLP) and speech. Task: Designing a robust, scalable infrastructure aligned with the company's vision, while implementing state-of-the-art machine learning models for specific applications. Actions: • Setting up an infrastructure using MLOps best practices. • Data collection, transformation and annotation. • Development and deployment of deep learning models (PyTorch). • Integration and adaptation of large language models (LLM). • Prototyping intelligent agents with LangGraph and LangChain. • Managing and optimizing computing resources. • Supervision of collaborators to ensure consistency of developments. • Documenting algorithm designs to facilitate knowledge sharing and ensure reproducibility. Result: Setting up an operational laboratory with optimized MLOps workflow, enabling rapid deployment of AI products. Stack : Python, Hugging Face, OpenAI API, LangChain, LangGraph, ZenML, PyTorch, Gradio, MLflow, Qdrant, Git, Github Actions CI/CD, Amazon Web Service (AWS), NumPy, Pandas, Pytest, Seaborn, Asyncio, Docker, Django, FastAPI, Tailwind CSS, Htmx, JavaScript.

Situation: The company wanted to develop artificial intelligence (AI) products specialized in natural language processing (NLP) for its customers. Task: Implementing and adapting state-of-the-art machine learning models for specific applications, while supervising projects and collaborators. Actions: • Design and development of Python algorithms for Named Entity Recognition (NER), exploiting Large Language Models (LLMs) such as BERT and RoBERTa. • Supervision of collaborators, ensuring their development and the quality of deliverables. • Advising on the choice and implementation of a high-performance infrastructure tailored to project needs. Result: Delivery of AI solutions tailored to business needs. Stack : Python, PyTorch, PyTorch Lightning, Docker, Git, Jupyter, Seaborn, NumPy, Pandas.

Situation: The laboratory wanted to analyze historical French archives to extract usable information. Task: Designing artificial intelligence (AI) algorithms to automate the analysis of ancient documents. Actions: • Development in Python of algorithms integrating optical character recognition (OCR), automatic transcription (ASR) and named entity recognition (NER). • Implementation of a pipeline integrating different technologies (OCR, ASR, NER) to automate complete archive processing. • Documenting algorithm designs to facilitate knowledge sharing and ensure reproducibility. Result: Delivery of tools for better understanding and use of period documents, facilitating historical research and the enhancement of archives. Stack : Python, PyTorch, Keras, Tensorflow, SLURM, Git, Scikit-learn, NumPy, Jupyter.

Situation: The company wanted to help documentalists in their work of annotating audiovisual data, where the growing volume of archives requires automation solutions. Task: Developing artificial intelligence (AI) diarization algorithms to automate speaker annotation. Actions: • Development in Python of diarization algorithms taking human intervention into account. • Active contribution to two Open-Source Python projects (S4D and SIDEKIT) specializing in speaker recognition and diarization. • Development of an evaluation metric to accurately determine the performance of human-assisted diarization AI algorithms. • Publishing results in the form of scientific articles and present them at international conferences. • Working in an international context in English, collaborating effectively with non-French-speaking colleagues and partners. • Documenting algorithm designs to facilitate knowledge sharing and ensure reproducibility. Result: Delivery of a tool to improve the annotation speed of audiovisual data, facilitating the work of documentalists. Stack : Python, Jupyter, Git, SLURM.