Collias, France · Available · Remote or hybrid
I spent fifteen years running operations teams across three countries. Then I started building the tools instead of specifying them. Everything below is live, in production, and built solo with AI.
Nicolas ChareilleCollias, France · FR / EN
I am not a developer, and that is the point. I am an operations person who learned to direct AI the way I used to direct a team: define the outcome, set the constraints, review the work, ship it.
At Google Ireland I led a Quality Engineering team building the data-quality pipelines that fed model training. At Dodax in Vienna I ran customer service for a €40M e-commerce business with 25 agents across two sites. I know what breaks at scale, and I know that an automation laid over a process nobody measured just produces the same errors faster.
Since 2025 I have shipped ten products and websites on my own: an AI phone receptionist with a number you can call right now, a film recommendation engine with payments wired, an AI-narrated fiction podcast that went from $368 to $7,059 a month in five months on about two hours of work a day. Different domains, one method.
Every item below is live and clickable. No mockups, no concepts, no case studies for things that were never built.
A studio that turns an author's finished novels into a serialised, distributed, monetised fiction podcast. The method is not theoretical: I built and run the show the numbers come from.
The back-office assistant for estate agencies. It reads 2,346 listings across the sector overnight, scores every one against each client's real criteria, and by morning the emails are drafted in the client's own language. Dictate your notes leaving a viewing and it returns a report ready to send to the vendor.
Pick a genre and a subgenre, and the AI ranks films by how well they fit your taste — each one scored, with a written reason for the match and where you can actually stream it. Skip, save or mark as seen, and the queue re-ranks around what you liked.
Try it — it's free right now.MovieMuse is in its founding period: everything is open, no card, no limits. The first 500 members keep Premium free for a year.
moviemuse.co ↗An AI phone receptionist for small French businesses. It answers, books into the calendar, confirms by SMS, and escalates the urgent calls to a mobile. Not a demo video — a real line you can dial.
A review platform for cities: what it is actually like to live somewhere, written by people who do. Built to answer a question I had myself while planning a move.
rate-my-city.com ↗My AI operations practice. The positioning is the proof: we do not hand over recommendations, we put our own products into production under our own name. Aura answers Aure-lab's phone.
aure-lab.com ↗A web studio for hotels, guesthouses and restaurants. Sites built to fill rooms and tables rather than win awards: local SEO, direct booking, and the Google presence most small hospitality businesses never get round to.
hospitalitedigital.fr ↗A family-support resource on hypospadias, a congenital condition my son was born with. Every medical claim on the site was reviewed and verified by Professor Kalfa, paediatric urologist. The one project here I would keep running whether or not it ever pays.
hypospadias.fr ↗Client work. Same stack, same method, shorter story.
The interesting part is not that AI wrote the code. It is that someone had to decide what to build, judge whether the output was good, and take responsibility for shipping it.
Volumes, contact reasons, time spent. An automation on top of an unmeasured process reproduces its errors faster. This comes straight from running support teams, not from a framework.
Written specs, hard limits, a defined stopping point. I brief AI the way I briefed a quality engineer at Google: the outcome, the constraints, and what "done" looks like.
Model output is a first draft from a fast junior. It gets read, questioned, and sent back. Most of the value I add is knowing which plausible-looking answer is wrong.
Live domain, real users, monitoring, backups, and someone on call — me. A prototype that never left localhost proves nothing about operating a system.
Fifteen years of operations before any of this. That is what makes the AI work land.