# Ivan de Murard > I've worked the floor. Now I build AI agents for the people who run it. AI Product Manager & Builder · Paris. Front of house at Les Grands Verres, La Grande Épicerie de Paris, Paris La Défense Arena, festivals and private events. Co-founded Sonor: 0→1 urban noise SaaS, incubated at Matrice, MVP shipped. 500+ users on 100+ sites and 7 C-level executives aligned at DomusVi, €1.9M of receivables tracked and 40% recovered at Bouygues Telecom IoT. Since November 2025 I design, build and evaluate AI agents that work the way a floor does: one decision a day, a person who answers, the outcome recorded next to it. Focus areas: agentic AI, boutique hotel operations, service design, SaaS, open data. - Site: https://ivandemurard.com - LinkedIn: https://www.linkedin.com/in/ivandemurard/ - GitHub: https://github.com/IvandeMurard - X: https://x.com/IvanMurard - Email: ivandemurard@gmail.com - Schedule: https://cal.com/ivandemurard/discussion-and-introduction - For agents (MCP server, mirrors, guarantees): https://ivandemurard.com/agents (markdown: https://ivandemurard.com/agents.md) - CV: https://ivandemurard.com/cv - CV PDF: https://ivandemurard.com/Ivan_de_Murard_CV.pdf - CV Markdown: https://ivandemurard.com/Ivan_de_Murard_CV.md ## Status vocabulary Every component status uses exactly one of these five labels. "Built" means the code runs, not that anyone uses it yet. Canonical source: https://github.com/IvandeMurard/Hospitality-Multi-agent-Architecture/blob/main/llms.txt - Built - deployed and exercised end-to-end on the target environment. - Shadow-mode - runs on real data, but no decision is delivered to a human on that basis. - Synthetic PoC - validated on fabricated data only; never met the real world. - Design - specified (ADR, schema, contract), not implemented. - Research - open exploration, no delivery commitment. ## Case studies - [Aetherix](https://ivandemurard.com/aetherix) (markdown: https://ivandemurard.com/aetherix.md): The F&B agent for boutique hotels: one recommendation a day in WhatsApp, the manager's reply, and the outcome recorded next to it. Built to improve with every day it runs. Built and evaluated on historical data. Staged on Fly.io, connected to the Apaleo sandbox. Ready for a 30-day pilot. - [Lore](https://ivandemurard.com/lore) (markdown: https://ivandemurard.com/lore.md): Voice AI mentor for aviation maintenance. Captures what a retiring technician knows and never wrote down, then gives it back to a junior on the shop floor by voice. Three modes: Capture, Query, Log. The procedure always wins, the senior's judgement always arrives attributed by name and month. Built in 24 hours at Speechmatics x AI Collective Paris, Feb 2026, developed solo since. Prototype: synthetic procedures, mock aircraft data, evaluation harness in CI. Stack: Speechmatics real-time STT, OpenAI GPT-4o, Backboard, OpenAI TTS, Next.js 14, TypeScript, Tailwind, Vercel. - [Anima](https://ivandemurard.com/anima) (markdown: https://ivandemurard.com/anima.md): The guest memory node of the Hospitality Multi-Agent Architecture. Memory that means no guest has to ask twice: it keeps what a guest said and what happened, so nobody asks twice, without watching. Under the hood: four memory layers with different lifetimes (working, episodic, semantic, segment), claims carrying a weight, a source and an age that decay unless renewed, and a scope gate so no call returns everything known about a person. It never decides; the orchestrator does, with a human validating. Synthetic proof of concept shipped at an Anthropic hackathon, June 2026. Never run on a real guest, deployment frozen until the data-protection impact assessment (DPIA) clears. - [Compass](https://ivandemurard.com/compass) (markdown: https://ivandemurard.com/compass.md): Every address has a story. Compass reads it before you sign. Open data turned into traceable, address-level context for commercial location decisions in Paris. APUR BDCom 2023 shopfront census · BODACC legal notices · INSEE Sirene · 85,418 units · every figure with its source, licence and date. Earlier census records remain withheld until their licence is cleared. Corpus and MCP server live; web demo on a frozen sample (17th arrondissement). Live app: https://paris-compass.lovable.app. MCP server: npx -y paris-compass-mcp. Code: https://github.com/IvandeMurard/paris-compass - [Sonor](https://ivandemurard.com/sonor) (markdown: https://ivandemurard.com/sonor.md): B2G SaaS that turned acoustic open data into a decision tool for cities. Taken from discovery to MVP; stopped in 2022 after discovery concluded no contract would clear. - [Tacet](https://ivandemurard.com/tacet) (markdown: https://ivandemurard.com/tacet.md): Headless Environmental Twin for hospitality, exposed via an MCP server. A 3D ray-tracing physics engine plus a dual memory store (idiosyncratic and Hive Mind, Research: no network to aggregate yet) turns street, weather and crowd signals into human-approved yield rules and Chain of Thought payloads for RMS, PMS and the orchestrator (Design). Compass belongs under Standalone projects. It is not a node, agent or part of the Hospitality Multi-Agent Architecture. It is agent-ready: its published, read-only MCP server gives AI agents the same figures, sources and confidence levels as a person sees on the map. The product contains no AI yet. Its six sections are One address, read in context; Product strategy (The problem, The bet, What it refuses, Against what exists); Where AI sits; Built for people and AI agents; How sure is it?; Where it stands. ## Field Notes - [Journal index](https://ivandemurard.com/journal) - [What is an LLM evaluation harness? Graders, protocols and closed loops for AI agents](https://ivandemurard.com/journal/harnesses-graders-closed-loops) (markdown: https://ivandemurard.com/journal/harnesses-graders-closed-loops.md) (2026-07-31): A working vocabulary for building evaluation into AI products from day one: case set, grader, harness, protocol, open loop, closed loop. Tested on Aetherix and Lore. - [A wrong forecast is absorbed. A wrong action reaches a guest.](https://ivandemurard.com/journal/agent-boundaries-aetherix) (markdown: https://ivandemurard.com/journal/agent-boundaries-aetherix.md) (2026-04-20): Prophet is a statistical forecasting library. Claude is a large language model. Keeping them as two components, rather than letting one model both predict and act, is the decision that shaped everything after it. ## Track record The complete CV, including Founder & Builder, Product & Customer Roles, Skills and Education, is available at https://ivandemurard.com/cv and as Markdown at https://ivandemurard.com/Ivan_de_Murard_CV.md. - 2024 to 2025, Product Manager at FORTIL for DomusVi: delivered the MVP; rollout planned for 500+ users across 100 care homes. - 2023 to 2024, Product Manager at ALTEN for Bouygues Telecom: data-driven dashboard recovered 40% of 1.9M EUR receivables. - 2022 to 2023, Customer Success Manager at Opendatasoft: 22 European city and metropolitan clients, 2 launches, zero churn. - 2020 to 2022, Co-founder & Product Manager at Sonor: 0 to 1 B2G SaaS, backed by Matrice and Banque des Territoires, 20K EUR in grants. ## Markdown representations Every document below is served as markdown at its own URL, identical in content to the page. - [Homepage](https://ivandemurard.com/index.md): hero, Aetherix flagship, Selected projects with the current status of each project (Aetherix WARN Sandbox, Lore GO Evaluated, Anima STOP Gated, Compass WARN Partial, Sonor DARK Stopped) - [Aetherix](https://ivandemurard.com/aetherix.md) - [Anima](https://ivandemurard.com/anima.md) - [Architecture](https://ivandemurard.com/architecture.md) - [Compass](https://ivandemurard.com/compass.md) - [Lore](https://ivandemurard.com/lore.md) - [Sonor](https://ivandemurard.com/sonor.md) - [Tacet](https://ivandemurard.com/tacet.md) - [A wrong forecast is absorbed](https://ivandemurard.com/journal/agent-boundaries-aetherix.md) - [What is an LLM evaluation harness?](https://ivandemurard.com/journal/harnesses-graders-closed-loops.md) - [Ivan de Murard CV](https://ivandemurard.com/Ivan_de_Murard_CV.md)