Freelancer profile translated to English.

Description

Generative AI today produces two things in abundance: impressive demos, and POCs that end up gathering dust. My job is the third: systems that run in production, are traceable, and can be entrusted with real decisions.

Specifically, I designRAGsystems for due diligence, documentary monitoring, or assisted writing, with knowledge graphs and anti-hallucination guardrails. I buildagentic architectureswhere specialized agents coordinate around tools, persistent memories, and multi-model routing (LiteLLM, MCP, PydanticAI, LangGraph), with the required observability in production. I also conductAI auditsand feasibility studies, and industrialize prototypes stuck at the POC stage.

I draw on 25 years of freelance software engineering, which have forged a dual skill set: in-depth expertise inLLMs(fine-tuning, structured outputs, evaluation, advanced RAG), and the versatility of a **senior developer comfortable across the entire stack**, from backend to frontend, including data, DevOps, and project management.

In parallel, I conduct active research inbrain-computer interfacesand develop open-source tooling for neuroscience. My work has led to publications in peer-reviewed journals, with some results advancing the state-of-the-art on open problems. This field requires real-timemachine learningon noisy signals, with severe latency constraints. A rigor that I then bring back to my clients.

At a time when code is becoming a commodity, value is shifting to the engineer who masters the entire chain, identifies real problems before they become costly, and knows when AI is the right answer and when it is not. If this is what you're looking for, let's talk: remotely, in French or English.

Industry field of expertise

Languages

English
Fluent
French
Native or bilingual

Workplace preferences

Remote only

Primarily works remotely

Confidentiel
AI Architect
TRANSPORTATION
October 2025 - January 2026 (3 months)
Paris, France
Due Diligence RAG System
Design and development of a RAG (Retrieval-Augmented Generation) system for automated analysis of large document corpora (data rooms, financial reports, legal contracts) for M&A due diligence operations.

Implemented Architecture
Multi-format ingestion pipeline (PDF, DOCX, XLSX, images) with OCR, semantic chunking, and structured metadata extraction
Hybrid indexing on LlamaIndex Cloud / LightRAG combining dense vector search and knowledge graph for entity resolution and multi-hop reasoning
Citation enforcement ensuring traceability of each response to its sources (page, paragraph, original document)
Hallucination detection through post-generation verification (NLI, RAGAS metrics) and cite-or-abstain policy
Agentic orchestration for complex queries requiring decomposition and cross-document synthesis
Deliverables
Fully functional product delivered as a ready-to-deploy Docker Compose stack, detailed technical architecture, volume scenarios (indexing, inference, storage costs), and integration documentation.

Stack:Python, FastAPI, LlamaIndex Cloud, LightRAG, PostgreSQL (pgvector), Neo4j, Redis, LLMs & embeddings, OCR, Docker Compose.

Keywords:RAG, due diligence, knowledge graph, GraphRAG, hallucination detection, source traceability, LLM, generative AI, NLP, multimodal document processing.
Artificial Intelligence RAG Python Knowledge Graph Vector Database
Multiples clients
Senior IT Consultant: AI/ML, Neurotech, Full-Stack & Security
December 1999 - Today (26 years and 5 months)
Over 25 years of freelance experience, covering a wide range of IT disciplines, for clients ranging from large institutions to startups and research labs.

AI & Machine Learning —Data engineering, supervised/unsupervised/reinforcement learning, deep learning, time series, knowledge graphs, embeddings, vector search, RAG, LLM integration and finetuning, agentic AI, evaluation, guardrails. Frameworks: Pydantic AI, DSPy, Smolagents, LangChain/LangGraph. Clients: UNESCO, ISAE-SUPAERO, Inclusive Brains, TTH, Splace.

Brain-Computer Interfaces & Neurotech —Real-time EEG/ECG/PPG/EDA acquisition and processing, DSP, ML pipelines, neurofeedback, motor imagery, evoked potentials, sub-millisecond synchronization. Author of Timeflux (open source) and Neurogate (hardware). Winner of the IEEE International BCI Competition. Clients: ISAE-SUPAERO, Tampere University, Banaras Hindu University, NeuroTechX, IRIS Intuition, Conscious Labs.

Web Development —Backend & frontend in Python, PHP, JavaScript, SQL/NoSQL; REST & JSON-RPC APIs; FastAPI, Flask, Vue.js, Angular, Ionic; Drupal, WordPress, WooCommerce; hybrid mobile apps; SEO; semantic web. Clients: UNESCO, Groupe Pasteur Mutualité, Campus France, Dassault Systèmes, EDF R&D, M6 Web, Fortis Banque, Région Île-de-France.

Cybersecurity —Vulnerability and penetration testing, code audits, system and network security, social engineering, intrusion detection. Clients: BPCE, École Militaire, Compagnie des Alpes, Clear Channel.

System & Network Administration —LAMP/LEMP, Docker, virtualization, CI/CD, server hardening, DNS, backup policies.

Telecom —VoIP architecture, call and SMS automation, IVR, CRM integration (Salesforce, Zoho, Sugar). Clients: Callr, SAGAC, Quotatis.
Web Development Cybersecurity Neurotechnology Machine Learning Artificial Intelligence
Timeflux
Neuroscience Engineer
BIOTECH
January 2017 - Today (9 years and 5 months)
I am the author ofTimeflux**, an open-source framework in **Pythonfor real-time acquisition and processing of biosignals, published in the proceedings of the 8th International BCI Conference, Graz, 2019. Designing and maintaining such a project requires cross-disciplinary expertise: modularsoftware architecture(execution graphs, plugin system),low-latency asynchronous programming**, **hardware integration(EEG, ECG, EDA, eye-tracking drivers),real-time protocols(LSL, ZeroMQ, OSC),applied neuroscience(BCI paradigms, EEG neurophysiology),signal processing(SciPy, Pandas, Xarray),machine learning(scikit-learn, Riemannian geometry), design ofuser interfaces(web, stimulus presentation), and community management for anopen-sourceproject.

Timeflux is used in production by neurotechnologycompanieslike Inclusive Brains, Conscious Labs, and Open Mind Innovation, as well as by internationalresearch institutionssuch as the French Institute of Aeronautics and Space (ISAE-SUPAERO), Lund University (BCI-HIL framework), Tampere University, and GIPSA-lab.

The framework has enabled several results **surpassing the state-of-the-art**. In 2022, the ASAP pipeline, based on Bayesian accumulation of Riemannian probabilities, significantly outperformed standard P300 classification methods on public datasets. In 2024, the StAR paradigm improved cVEP accuracy from 65.6% to 96.3% with 88s of calibration, and reached 97.5% online on an 8-electrode dry EEG, paving the way for usable BCIs outside the lab.

I have also given over 50 talks and workshops on these topics, in both academic and industrial settings.
Machine Learning Neuroscience Python EEG Signal Processing