I design and operate production GenAI systems, from agentic workflows and RAG to evaluation, observability, security and cloud infrastructure.

Independent Data & AI consultant and founder based in Paris, with 20 years across IT, data and AI.

View consulting work · Connect on LinkedIn

• grafana-llmops-forge: a dependency-free forge that discovers a Grafana deployment and generates, deploys and visually verifies dashboards for LLM observability, FinOps, agent traces, quality and governance.
• generative-engine-monitor: measures brand visibility in parametric and grounded AI answers with bootstrap confidence intervals, explainable scoring and zero-cost replay.
• tierdecay: a self-distilling model router for AI coding CLIs that learns task-class routing from execution feedback, using Markdown only and no proxy.
• harnessmeter: an offline profiler for CLAUDE.md instructions, skills, subagents and MCP schemas across Claude Code, Codex and Antigravity, with no API keys or network access.
• ftp-deploy-mcp: an MCP deployment server for FTP, FTPS and SFTP with multi-server support, path isolation, read-only mode, dry runs and local credentials.
• claude-inc: a Claude Code company framework with peer CEO and CTO executives, eight department charters and 54 skill manuals; selected work runs through the host's agents and permissions, while local project state records tasks, artifacts and reviews.
• open-fullscreenshot: a Chrome Manifest V3 capture extension for whole pages, viewports, elements and regions, with no host permissions, network calls or telemetry.

• AgentShield: a merged scanner fix that discovers and classifies root .mcp.json files so MCP configurations reach security rules, with regression coverage; all 16 review threads were resolved.
• Claude SEO: improved JSON-LD detection and removed template-expression false positives; the maintainer confirmed it was cherry-picked with authorship preserved into v2.3.0.
• LibreChat: a merged documentation fix that replaced a machine-specific agents source path and restored CLAUDE.md/AGENTS.md parity for auth-cache guidance.
• Grafana Plugin Tools: two scoped merged contributions: signed-commit contributor guidance and a Node 24 .nvmrc scaffold update.

• Production GenAI architecture and the path from prototype to operation
• Repeatable evaluation at component and system level
• Observability for quality, cost, latency and failure analysis
• Security, data boundaries and controlled tool access

A focused stack for agentic orchestration, evaluation, serving and cloud-native operations.

LangGraph	CrewAI	Hermes Agent	OpenAI Agents	MCP	Claude Code
RAGAS	Langfuse	OpenTelemetry	Hugging Face	vLLM	LiteLLM
Docker	Kubernetes	Terraform	Prometheus	Grafana	PostgreSQL
LangChain	Codex	PyTorch	Ollama	FastAPI	Grok
AWS	GCP	Azure AI	pgvector	Redis	Qdrant
Python	TypeScript	JavaScript	Bash	Git	GitHub Actions
Gemini	Qwen	Z.ai	DeepSeek	Mistral AI	OpenRouter

• Start with the problem, constraints and failure modes.
• Treat evaluation and observability as product requirements.
• Minimize privileges and data exposure; make consequential actions explicit.
• Prefer simple, reviewable components and measurable automation.

beguel.com · LinkedIn

French is my native language; I work professionally in English and Spanish.