Making a hotel's own numbers answerable in a sentence.

A hotel already generates everything it needs to make good decisions. The trouble is where that information lives: the property management system, the revenue management system, the rate shopper, the ad platforms, the analytics, the point of sale, the review feeds. Each is a separate login with a separate export.

RoomAngel's own framing is that revenue managers spend fifteen to twenty hours a week simply compiling data across those systems before any decision gets made. Marketing and revenue teams end up working from different pictures of the same week. Leadership has no single view of how an asset is actually performing.

Hive's answer is to unify the feeds and put a conversational interface on top, so the question a manager actually has - why did last weekend underperform, what is our position against the comp set - can be asked directly rather than assembled by hand.

I work across the platform rather than in one layer of it: backend services and the frontend that sits on them, which for a forward deployed role means being close enough to the customer to see which gaps actually matter.

A large part of the job is integration. Hospitality runs on established third-party platforms, and each property group arrives with a different combination of them. I built integrations against those platforms with multi-tenancy throughout, so one deployment serves many properties without their data or configuration bleeding across.

On the product surface, I delivered the features that let a hotel interrogate its own revenue and revenue-centre performance through the natural-language interface - turning a question typed in a chat box into a grounded answer drawn from the connected systems.

I also built agents that handle the repetitive work inside that interface. Where a task was being done the same way every time, it became something the system does on its own rather than something a person retypes.

The part I would point to first is not a feature. It is the harness the team builds with.

I authored the custom agents, skills, hooks and orchestration that let a single command carry a feature through the backend, the frontend and the infrastructure together, with the conventions of the codebase already encoded rather than re-explained each time.

The effect is measurable in delivery rather than in a demo. Feature work that took weeks now lands in two to three days, at 98% accuracy, with no rework cycle afterwards. That last part matters more than the speed: fast delivery that comes back for repair is not fast.

The agent harness paid for itself, but it was built while shipping rather than before. If I were starting again I would encode the conventions earlier, when there were fewer of them to capture and less existing code disagreeing with them.

The other thing I would change is how early the integration surface got a shared shape. Each platform arrived with its own model and the common ground only became obvious after several were in. Building the third one is when you learn what the first two should have looked like.