Freelancer profile translated to English.
Back to original languageAbout Thomas
How to deploy an autonomous AI Agents system?
One that is reliable, auditable, executes business tasks in production while respecting cost constraints? This has been my favorite R&D area since the public access to the first building blocks of intelligence represented by generative AI.
My experience:
🏭 Data Scientist at Qualcomm — Computer Vision & XR
In an industrial production environment: models running on constrained hardware, at scale, with the robustness requirements of a semiconductor player. This is not prototyping, this is deliverable engineering. I am applying this edge/hardware rigor to agentic systems.
🤖 Autonomous AI Agents & Cognitive Systems
The core of my practice. I design multi-agent architectures that reason, decide, and execute:
- Orchestration of complex cognitive workflows with rigorous state management and typed guardrails;
- Frameworks: PydanticAI (structured outputs, strong validation);
- Tool interoperability via Model Context Protocol (MCP) — the standard that decouples agents and capabilities;
- Multi-model architecture: the right model for the right task, to respect cost constraints.
☁️ Scalable AI Infrastructure & Edge MLOps
An agent without robust infrastructure is a demo. I deploy for production:
- Kubernetes clusters, Docker containerization
On-Premise / Cloud / hybrid deployment, sovereignty, and cost control (FinOps);
- Edge AI on specific hardware, inference optimization.
🔍 Advanced RAG & Data Engineering
The foundation that anchors agents in your data:
- Agentic RAG, vector databases, graphs, dynamic context injection.
📍 Based in Bordeaux. Full-remote or hybrid missions (occasional travel possible).
💬 Do you have an autonomous system to design, secure, or scale? Let's talk architecture.
French
Native or bilingual
English
Fluent
Remote only
Primarily works remotely
Experience
- Agentic Systems EngineeringIndependent R&D — Multi-Agents & InfraTECHSeptember 2022 - Today (3 years and 9 months)Bordeaux, FranceDesign of an autonomous, self-replicating agent swarm — end-to-end production-grade agentic architecture.🧠 Cognitive LayerTyped agents using PydanticAI (validated structured outputs, reliable LLM calls), custom asyncio runner with a priority message loop and preemption mode — explicit management of context, error recovery, and autonomous thinking cycles (idle / cron).🔌 InteroperabilityExposing all capabilities via decoupled MCP servers (filesystem, sandboxed execution, websearch, Telegram, swarm orchestration), fractal architecture where each agent inherits only the tools explicitly delegated by its parent — guarantee of isolation and least privilege.🧬 Memory & ContextFour-level memory system inspired by cognitive sciences (Tulving) — working memory (context window), episodic-filter (Redis, TTL 24h), episodic-swarm (TimescaleDB hypertable, append-only journal), and semantic (pgvector + text search) for dynamic knowledge injection across sessions.⚙️ InfrastructureAgents containerized in Docker/Kubernetes Pods, dynamic spawning via MCP deploy (the swarm deploys itself), stateless daemon ensuring high availability of the main agent, role-based routing via Redis Sorted Sets with priorities and preemption — multi-model architecture (lightweight model for routing/delegation, heavy model for reasoning) to control inference costs.✅ Validated ResultStable autonomous swarm on multi-step tasks with hierarchical delegation, complete observability (structured event logs, LLM traces per tool call, queue metrics), reproducible and redeployable On-Premise.
- QualcommData Scientist — Computer Vision & XRTECHDecember 2022 - Today (3 years and 6 months)Bordeaux, FranceDesign and deployment of computer vision models for XR, AR, and IoT use cases, under industrial production constraints: latency, robustness, embedded hardware.🎯 Demo & IntegrationModel demos & fluid integration with AI agents for concrete use cases (AI assistant for Smart Glasses, live Stable Diffusion, etc.).⚡ Model OptimizationQuantization, fine-tuning to meet the compute/memory budgets of the target hardware while improving quality.🧠 CV AlgorithmsDetection, segmentation, pose estimation — designed for the improvement of our Deep Learning models.🏗️ NASDAQ-100 Engineering RigorContainerization, scalability, versioning, reproducibility, validation.
Recommendations
Be the first to recommend Thomas
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
- Master's Degree | AI EngineerOpenClassrooms2022
- Master's Degree | Computer Science EngineerEPSI, the computer engineering school2021