Session abstract
What you’ll learn
Some of the best products don’t start on a roadmap, they begin as ideas in internal innovation challenges. This talk tells the story of how one such project evolved into the Hilton AI Planner, a generative AI powered digital concierge now live on hilton.com and already gaining attention from major travel publications. Currently in beta and expanding to more users throughout 2026, the Hilton AI Planner uses conversational intelligence to deliver real time, curated recommendations, helping travelers navigate every stage of trip planning with a more intuitive, frictionless experience. This was new territory for one of the world’s largest hospitality brands, and it required deep collaboration across engineering, SRE, SDET, product, UX, and security. Developers played a key role, not just building features, but partnering across disciplines to help shape core product decisions and define what this new category of experience could and should be. We’ll go under the hood on the technical architecture, how we leveraged cloud based foundation model orchestration to power conversational intelligence, built agentic workflows that combine Hilton’s rich hotel portfolio data with real time responses, and grounded LLM outputs in structured hospitality data at scale. We’ll also share lessons from key pivots, including a major shift mid build from a managed agentic framework to a self hosted solution, and why continuous evaluation has become a core team discipline in a rapidly evolving ecosystem. Equally important is how we protect and build trust with our users. We’ll cover how we implemented guardrails to ensure safe, reliable, and brand aligned responses, alongside enterprise observability tooling to trace agentic workflows and catch regressions. We’ll also explore how LLM as a judge frameworks help us evaluate conversational quality at scale, because when your product is a conversation, your testing strategy has to be just as intelligent. The beta is live. The rollout is 2026. This is what Hilton’s test and learn approach to generative AI looks like from the inside, and how it continues to evolve in production.