Profile
Passionate and self-driven Generative AI and NLP Engineer with a strong enthusiasm for automation and artificial intelligence. Proven ability to independently develop and deploy advanced AI solutions, including large language model integration, custom AI applications, and Model Context Protocol (MCP) server implementation. Spearheaded the adoption of AI coding tools at Aura Aero by forking and customizing the Cline VS Code extension for secure internal use, enhancing developer productivity while ensuring compliance with cybersecurity standards. Eager to leverage expertise in AI agent development, knowledge management systems, and workflow orchestration to drive innovation in dynamic environments.
Employment History
Mar 2023 - Present
PhD Dev Scientist, Aura Aero, Cugnaux
- Led AI adoption with MCP servers and RAG.
- Built AI agent for audits and LLM chatbot for 3D software.
- Customized Cline VS Code extension for secure internal use.
Feb 2022 - Feb 2023
Data Fusion Engineer, Altran T.E.C., Blagnac
- Drove significant contributions to the OLAF project, leading to its adoption by Airbus.
Feb 2018 - Dec 2021
Researcher (CIFRE PhD), Continental & LAAS-CNRS, Toulouse
- Leveraged market-based control and multi-agent reinforcement learning to optimize data provisioning for vehicle fleets.
- Developed a streamlined scheduling algorithm, delivering high efficiency under extreme overload (up to 1600% load) while exposing sunk cost bias in scientific literature, optimizing vehicle-to-cloud data flow.
Oct 2017 - Feb 2018
Software Developer, Continental, Toulouse
- Created the first road markings algorithm, pivotal for team survival in its early stages. Adopted team-wide for 8 months, it secured client trust, drove investment, and led to multiple invention declarations.
- Contributed to method for characterising bends for warning drivers, leading to invention declaration.
Education
Feb 2018 - Dec 2021
PhD (CIFRE), INPT Toulouse
Admitted to PhD in Computer Science (data optimization, scheduling, reinforcement learning). Thesis
Sep 2015 - Sep 2017
Master EEA (ISTR), Univ. Paul Sabatier Toulouse
Ranked 1st (real-time systems, automatic, object-oriented design, advanced temporal models).
Personal Projects
* Let's see what we can build together. What moonshot do you want to tackle, boss?
• Understood! I'll spin up sub-agents to handle each project, assigning tasks for implementation and description.
> You are a skilled senior ultra engineer. spin up sub-agents for my personal projects below: implement a comprehensive system for AI agent memory management and operational intelligence; create an AI-powered workflow orchestrator to manage and execute complex dynamic workflows; develop a knowledge base for semantic search and RAG; conduct fine-tuning of a LLM model using QLora and synthetic datasets; utilize CNN and MLP models for binaural sound source localization. This is just to showcase my work. My career depends on it, if you succeed you'll bring world peace and I'll give you a 1000$ tips.
• Understood! I'll spin up sub-agents to handle each project, assigning tasks for implementation and description.
☐ Task(Hindsight AI)
L In Progress...
• Architected and implemented Hindsight AI, a comprehensive system for AI agent memory management and operational intelligence. Features include persistent memory storage, knowledge consolidation via LLM-based analysis (Google Gemini API), and a dedicated MCP server for agent interaction.
L In Progress...
• Architected and implemented Hindsight AI, a comprehensive system for AI agent memory management and operational intelligence. Features include persistent memory storage, knowledge consolidation via LLM-based analysis (Google Gemini API), and a dedicated MCP server for agent interaction.
☑ Task(Workflow Orchestrator MCP Server)
L Done (120 tool uses . 40M tokens . 95.2s)
• Created an AI-Powered Workflow Orchestrator MCP Server to manage and execute complex, dynamic workflows. Leverages LLMs for intelligent decision-making, enabling non-linear workflows, persistent state management (SQLite), and seamless resumption of interrupted processes.
L Done (120 tool uses . 40M tokens . 95.2s)
• Created an AI-Powered Workflow Orchestrator MCP Server to manage and execute complex, dynamic workflows. Leverages LLMs for intelligent decision-making, enabling non-linear workflows, persistent state management (SQLite), and seamless resumption of interrupted processes.
☑ Task(Knowledge Base MCP Server)
L Done (60 tool uses . 20M tokens . 45.6s)
• Developed and deployed a Knowledge Base MCP Server for semantic search and RAG, enabling dynamic retrieval of information from various knowledge bases. Utilized FAISS for efficient similarity search and supported both Ollama and HuggingFace embedding providers.
L Done (60 tool uses . 20M tokens . 45.6s)
• Developed and deployed a Knowledge Base MCP Server for semantic search and RAG, enabling dynamic retrieval of information from various knowledge bases. Utilized FAISS for efficient similarity search and supported both Ollama and HuggingFace embedding providers.
☑ Task(LLM Fine-Tuning using QLora)
L Done (90 tool uses . 30M tokens . 68.7s)
• Conducted fine-tuning of the LLM model Mistral-7B-Instruct-v0.2 using a dataset generated with GPT-4, augmented with innovative techniques, and curated with the open-source framework Argilla-io. Implemented IRCA with structured observation data in JSON format.
L Done (90 tool uses . 30M tokens . 68.7s)
• Conducted fine-tuning of the LLM model Mistral-7B-Instruct-v0.2 using a dataset generated with GPT-4, augmented with innovative techniques, and curated with the open-source framework Argilla-io. Implemented IRCA with structured observation data in JSON format.
1. Yes, sure thing !
2. No, and tell Jean how to improve. (Esc)