AI and Backend Developer · Preeminent Technologies · Jul 2024 - Jan 2025 Space Beacon, a voice-first AI travel platform, integrating 2M+ destinations and generating $425K in year-one GMV with zero marketing spend
Space Beacon is Preeminent Technologies' AI travel planning and booking platform. Instead of the standard pattern of tabs, filters, and comparison shopping across separate flight and hotel sites, a traveler tells Beacon's voice assistant where they are going, when, and how they like to travel, and the assistant searches, compares, and books on their behalf.
I joined as a developer on the AI and backend team. I want to be direct about scope: I was building, not setting product direction. The decisions below are the ones I influenced or executed as an engineer with product judgment, not calls I owned as a PM would.
Trip planning is scattered by design. Flights live on one site, hotels on another, activities on a third, and a traveler is left holding the comparison work themselves across all of it. That fragmentation is the default experience nearly every travel platform on the market still has to work around, because it comes from how the travel supply chain is structured, not from any one company's product choices.
Space Beacon's founding thesis was that a single conversational interface, voice or chat, could absorb that comparison work instead of handing it to the traveler. Say where you want to go and how you like to travel, and the assistant does the searching and cross-referencing that would otherwise mean a dozen open tabs.
I want to be honest about this section rather than reconstruct research I was not part of. As a developer, I was not the one running user interviews or validating the founding thesis. What I can speak to is the market condition the thesis responds to: multi-app, multi-tab travel comparison is a well-documented source of friction in travel booking generally, and a conversational interface that removes the comparison burden was the product's bet on solving it. I built against that thesis. I did not originate it.
Coverage over control: integrating Viator instead of building direct supplier relationships. For an MVP, the priority was giving travelers as many real options as possible across as many destinations as possible. Viator's enterprise API let us map more than 2 million destinations into the platform immediately, with real-time flight, hotel, and availability data from multiple providers in a single search. Building that coverage through direct airline and hotel integrations one at a time would have taken far longer and left early users with a thin, unconvincing set of options. The trade-off was accepting a partner's data and commercial terms rather than owning those supplier relationships directly, in exchange for speed and breadth at launch.
Investing backend effort in scale and security before feature breadth. I built the platform's cloud-based backend APIs on .NET 8, C#, and Azure. With enterprise travel data flowing through the system from multiple providers, the priority was an architecture that could handle real travel-data volume securely and reliably, ahead of expanding what the assistant could do. A voice assistant that books travel has to be trustworthy with real transactions before it can be expansive.
Automating the fine-tuning pipeline rather than hand-tuning the model. Working with Azure OpenAI and big data tooling, I helped build automated workflows for training and fine-tuning the LLMs behind search and recommendations. Automating that pipeline, instead of tuning models by hand for each improvement, was what let the team iterate on search precision and recommendation quality repeatedly rather than treating a tuning pass as a one-time event.