# Ivan de Murard, AI Product Manager & Builder, Paris

AI Product Manager & Builder · Paris

## Front-of-house, co-founder, product manager. Now I build AI solutions

Since 2025, I design, build and evaluate AI workflows for operations, bridging operations, tech, and real business value.

## 01 · Flagship: Aetherix

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.

[How it decides, and when it refuses](https://ivandemurard.com/aetherix#what-it-does-today)

Built and evaluated on historical data. Staged on Fly.io, connected to the Apaleo sandbox. Ready for a 30-day pilot.

Hero: Aetherix · WhatsApp · mock-up. Three days from a sandbox hotel (synthetic data, real services, seed 42): https://github.com/IvandeMurard/Hospitality-Multi-agent-Architecture/tree/096509806333661409ab682b7d9e498ca7a67bcd/demo/closed-loop

### Day 1 · Who's right?

- Evening: covers expected 101. Why 101: usual level 87; hotel 98% full +9, New Year's Day +8, season +7, weather −6; other factors −4. High confidence: 94–109 covers on 4 nights out of 5. Staffing: 7 servers, 2 hosts, 3 runners, 5 kitchen staff, on the hotel's own rules (usual night: 5, 2, 2, 3).
- Manager (WARN): I'd say 79 covers instead. The simulated manager averages the last 4 Thursdays: 69, 81, 72 and 95.
- Next morning (GO): Actual: 95 covers. Forecast 101: 6% off, inside the 94–109 range. Your estimate 79: 17% off.

### Day 10 · It refuses

- Evening: covers expected 85. Why 85: usual level 85; Saturday +17, hotel 59% full −14, weather −5; other factors +2. High confidence: 78–92 covers on 4 nights out of 5. Staffing: 6 servers, 2 hosts, 2 runners, 4 kitchen staff, on the hotel's own rules (usual night: 5, 2, 2, 3).
- Manager (GO): OK
- Next morning (STOP): Yesterday's POS figure (900 covers) is implausible. It was discarded and will not feed learning. Could you check the till export? (A corrupted till export injected by the demo to test the guard: the figure exceeds what the restaurant can seat.)

### Day 17 · It sees it coming

- Evening: covers expected 82. Why 82: usual level 84; Saturday +17, hotel 64% full −11, weather −6; other factors −2. High confidence: 75–90 covers on 4 nights out of 5. Staffing: 6 servers, 2 hosts, 2 runners, 4 kitchen staff, on the hotel's own rules (usual night: 5, 2, 2, 3).
- Manager (GO): OK
- Next morning (WARN): Actual: 92 covers. Forecast 82: 11% off, outside the 75–90 range. That is the third under-estimate in a row: demand seems to be shifting (a new contract? a recurring group?). The weekly recalibration will take it into account.

## 02 · Selected projects

### Lore (GO · Evaluated)
Evaluation harness · 64 cases · pure-function graders · hard safety tier.
Captures the tacit knowledge a retiring technician never wrote down and returns it to a junior on the shop floor by voice, always behind the procedure.

### Anima (STOP · Compliance Gated)
Synthetic PoC. Architecture ready, deployment frozen until the data-protection impact assessment (DPIA) clears.
Hotels forget their guests: what they know sits in one person's head or in scattered tools. Anima keeps only what guests chose to share, and uses it to adapt the stay.

### Compass (WARN · Partial)
APUR BDCom 2023 shopfront census · BODACC legal notices · INSEE Sirene · 85,418 units · every figure with its source, licence and date.
Every address has a story. Compass reads it before you sign.

### Sonor (DARK · MVP shipped · stopped 2022)
0→1 with four co-founders, plus a developer and data scientist: discovery, MVP, grants, commercial discussions with cities. No contract closed; we stopped in 2022.
It taught me to ship one use case before ten, the rule Aetherix is built on.

## 03 · Field Notes

The decisions behind the products. Written to be reused. [All notes](https://ivandemurard.com/journal)

- [Keep the forecaster and the actor apart.](https://ivandemurard.com/journal/agent-boundaries-aetherix) One model that predicts and acts looks elegant on a slide. Why I split Prophet and Claude in two, and everything that single choice decided next.
- [Harnesses, graders and closed loops](https://ivandemurard.com/journal/harnesses-graders-closed-loops) Agents rarely fail loudly. They drift, quietly, until a customer notices. How to find out first: graders, harnesses, and the gaps I have not closed yet.
- [Whose truth is the ground truth?](https://ivandemurard.com/journal/whose-truth-is-the-ground-truth) Your agent learns from something: a forecast, a manager's yes, or what really happened the next morning. Pick the wrong one and you build a very confident yes-man.

## 04 · Contact

Bridging operations, technology and real business value

Let's build the solution your team needs.

Open to AI Product Manager and Product Builder roles.

- Book a call: https://cal.com/ivandemurard/discussion-and-introduction
- Email: ivandemurard@gmail.com
- Or ask your AI about me: Perplexity (https://www.perplexity.ai/search?q=...), prefilled with: "Read https://ivandemurard.com/llms.txt and summarise what Ivan de Murard has built and shipped, and which roles he would be strong in. Cite the pages you used." A "Copy prompt for Claude or ChatGPT" button copies the same prompt.