Turning a travel brief into a voice-first AI trip planner
SpaceBeacon — Building a Voice-First Travel Platform
Context
SpaceBeacon was building a voice-first travel platform that let users describe where they wanted to go and have the system do the research for them. I worked across the AI and backend layers, helping connect user intent, structured travel data, and a conversational flow that felt fast and useful instead of robotic.
The problem
Travel planning had become a multi-tab workflow: flights, hotels, neighborhoods, timing, seasonal trade-offs, and pricing all lived in separate places. The product needed to collapse that work into a single conversational experience while still surfacing trustworthy travel options.
What I did
- Built the backend plumbing for a voice-first trip-planning flow, turning user requests into structured trip criteria and action items.
- Worked on the AI orchestration layer that interpreted destination intent and mapped it to the right data sources and task flow.
- Helped connect travel data and recommendations into a workflow that reduced the number of manual steps users had to take.
- Shipped a faster decision loop so the product felt like an assistant rather than a search box.
The decision I’d defend
I pushed the product toward a “tell me where you are going” workflow rather than a generic chat experience. If the system understood the trip goal and the user context, it could carry the work of collecting and synthesizing the options instead of forcing the user to do the tabs themselves.
Outcome
The experience was designed to make travel planning feel like a guided copilot: ask once, get a structured answer, and move forward without the usual search fatigue. The core takeaway was simple — the value was not in adding more screens, but in removing the number of tabs a trip decision required.
What I'd do differently
I would have pushed harder on evaluation and retrieval quality earlier. A voice-first travel assistant is only trustworthy when the underlying data and recommendations are consistent, and that part benefits from explicit quality checks before it becomes a user-facing product default.