The booking workflow assumed hotels were passive. We built a marketplace that made them active.

Cerbex connects travel brokers, including tour operators and destination management companies, with hotels for group bookings. A broker may manage a multi-city itinerary, source a hotel for each stop, and collect competing rates before choosing one.

The broker side had been built and refined over time. It handled trip creation, hotel search, requests, and bookings. The hotel side was barely there. Hotels received email notifications, replied in different places, and had no single view of their pipeline, offers, or confirmed bookings.

My brief was to build the hotel product from scratch and modernize the broker side with AI-native tools. I also redesigned the small request page hotels use to submit offers without logging into the full app. It needed to fit the new booking flows. Taken together, the work changed the relationship between the two sides of the marketplace.

Group hotel booking usually starts with a broker choosing hotels they already know, sending a rate request, and waiting for replies. A hotel sits idle until someone contacts it. If a broker does not know the hotel, it never enters the pipeline.

Conversations between the CEO and his clients exposed the weak assumption in that model. Brokers want the right hotel for a group and a price point, rather than only the hotels they already know. Hotels have empty rooms on dates that match groups they have never heard about. The information was there. The market was not.

The marketplace lets hotels see every group coming to their area, rather than only the groups that requested them directly. They can choose which ones to bid on. Brokers get more options, and hotels get a way to find demand instead of waiting for it.

I kept picturing a reservations manager in the middle of a shift: phone calls, front-desk questions, and too many browser tabs. They do not get a quiet hour to study a dashboard. They need to triage in thirty seconds.

One question shaped the Hotel Center: what decision does this person need to make in the next thirty seconds?

Before the AI Trip Extractor, a client, often a smaller Asian travel agent, sent a trip description in whatever format was convenient: email, PDF, Word document, or pasted text. The broker then opened the itinerary builder and re-entered every destination, date, room split, and budget. A five-city tour with ten room categories took real time, especially when the same tour ran several times a year.

The AI Trip Extractor handles that conversion. A broker uploads a PDF, Word file, Excel sheet, CSV, or plain-text document, and the system extracts the itinerary into the builder. A series of trips on different departure dates can come from one document instead of being created one at a time.

Brokers reported a 10× improvement in trip-creation productivity. The gain came from removing the conversion step, not from making manual entry slightly faster.

The first Hotel Center exposed the booking lifecycle as brokers see it: request, offer, booking, confirmation. That matched the system. It did not match how hotels worked.

A reservations manager does not track groups they rejected. The broker needs to know that a request was sent. The hotel needs to track the groups it engaged with, which means the groups it made an offer on. Putting "requested but no offer made" into the hotel's pipeline added noise from the broker's mental model.

The marketplace made the distinction more important and easier to resolve. Once hotels could bid on any group in their area, the shared question became did we make an offer? A direct request and an open-market group remain separate entities in the backend. In the hotel UI, both become a group the hotel can offer on or has already offered on.

I did not erase the distinction. A broker's direct request still signals a prior relationship or a strong fit. It appears in a dedicated "Requests" section, as a "Requested" label where it matters, and in alerts that match the emails hotels already receive. The information is there without becoming the main way to navigate the product.

Reconciling that model took product, design, and coding work. The backend keeps requests and market groups separate, while the hotel UI presents a single working surface. The hard part was hiding the backend model without losing information the hotel needed.

I was product lead, designer, and co-developer across the Python, Postgres, and Next.js stack. I worked from strategy and roadmap decisions through Figma and production code. Cursor and Claude Code handled much of the implementation under my direction and review.

That matters because I could move from a design decision to a working interface and backend integration in the same session. When you are the person who will build the thing, the design is more honest. You avoid elegant ideas that are painful to implement, and you can resolve backend consequences while the decision is still fresh.

The Trip Extractor, the marketplace, and the new request and offer model all needed quick passes between product thinking, design, and code. Keeping that loop short made this scope possible for a team this small.

The Hotel Center is live with its dashboard, booking pipeline, market, team management, offer, and confirmation flows. The broker tools are in production too. The hotel product is rolling out to an initial group of hotels, with structured feedback coming next.

I would validate the hotel-side mental model earlier, before designing the full pipeline view. The idea that hotels do not track requests they never answered sounds obvious now, but it only surfaced while working through the full lifecycle. A rough test with one reservations manager would have caught it sooner.

I would also spend more time on the transition between the market and the pipeline. For now, the dashboard keeps the hotel's current bookings and the open-market feed in separate sections. As the market grows, that balance will need to change. The architecture can support it; the UI will need to catch up.

The marketplace is the part I am happiest with. It challenged an assumption the industry takes for granted: brokers choose hotels, and hotels wait. Here, both sides can act. That matters more than any individual feature.