Functional DemoAI Platform

TravelOS

An AI travel planner that turns loose trip desires into a structured, place-verified day-by-day itinerary.

What it solves

People describe trips as compressed signals ("a relaxing week somewhere warm") rather than structured plans, and turning that into a coherent, time-sequenced itinerary means juggling destinations, experiences, budget, and logistics by hand. The spec frames the goal as inferring what a traveler actually wants and producing a usable, structured travel plan from ambiguous input.

Built for Individual travelers planning personal trips.

Overview

TravelOS takes a destination and traveler intent and uses Claude to generate a ranked, time-sequenced itinerary, then optionally verifies each location through Google Places to add coordinates, hours, map links, and travel-time estimates. Trip cover photography is pulled from Unsplash when configured, and booking deep-links carry optional affiliate/partner referral codes. Sign-in is handled by email magic links or Google OAuth, and every AI generation is bounded by per-user cooldowns and daily spend caps because each run is a real model call. Optional integrations degrade cleanly to nothing when their keys are unset.

What this demonstrates

Implements AI itinerary generation grounded against a real places API, with per-user rate limiting and daily spend caps to control model cost, and a fixture-based offline planner for testing. Emphasizes safe production behavior: secrets/PII redaction before errors leave the app, a preflight config checker that fails when env and code drift, and each optional integration failing silent rather than breaking the page.

What makes it different

Most travel apps search listings. This one reconstructs what you actually want to experience, eliminates everything that doesn't fit, and can execute and adapt the trip on its own.

Users input signals, not plansDecode ambiguous, emotional requests into structured travel intent.
Preferences are relationships, not valuesAn evolving 'Travel DNA' graph, not static filters.
More options means worse decisionsAggressively eliminate poor options before ranking begins.
Rank by meaning per unit of timeNot by price or popularity.
Simulate futures, then chooseBuild competing full-trip realities and select the best.
Intelligence without governance is chaosA master orchestrator arbitrates conflicts between subsystems.
Coherent autonomy over maximum intelligenceAligned agents acting as one system, across defined autonomy levels.

Core engines · 15

Intent Inference & Emotional Modeling Engine
Turns incomplete, emotionally-driven input into machine-actionable travel intent.
Preference Graph & Travel DNA Engine
Builds a persistent, evolving model of the user's travel identity.
Global Travel Market Universe Engine
Unifies flights, hotels, and activities into one queryable travel universe.
Multi-Layer Filtering & Elimination Engine
Reduces millions of options to a small set of aligned candidates.
Experience Density & Value Scoring Engine
Ranks experiences by human value and quality per unit of time.
Trip Simulation & Parallel Outcome Engine
Generates competing full-trip realities and selects the optimal timeline.
Itinerary Compression & Flow Optimization Engine
Structures selected experiences into a frictionless, time-based journey.
Booking Orchestration & API Integration Layer
Executes real bookings across external APIs with transactional safety.
Real-Time Trip Adaptation Engine
Monitors and re-optimizes an active trip live as conditions change.
Budget Optimization & Financial Allocation Engine
Allocates spend to maximize experience value per unit of currency.
Master Control Orchestrator & Autonomy Governance Engine
Arbitrates conflicts between subsystems and governs autonomy levels 0–4.
Security, Permissions & Data Governance Engine
Enforces access control, data governance, and legal/financial compliance.
Production Observability & Operations Engine
Keeps the system healthy, observable, and controllable in production.
System QA Validation & Production Readiness Engine
Detects contradictions across modules and gates production readiness.
API Contracts & System Interfaces Engine
Defines schemas and contracts for internal and external integrations.

The intelligence loop

Decode user intent from ambiguous signals
Model the user's evolving Travel DNA
Unify global travel supply into one queryable universe
Eliminate poorly-fitting options before ranking
Score remaining experiences by value per unit of time
Simulate competing full-trip realities and pick the best
Compress the winner into a clean, flowing itinerary
Execute bookings across external APIs
Adapt the live trip to disruptions in real time
Optimize budget allocation for experience ROI

How it is built

A pnpm monorepo: a Next.js web app plus a shared trip-core planning package. Data lives in PostgreSQL through Prisma (Neon pooled connection in production), authentication is Auth.js v5 with a Prisma adapter, the server calls Anthropic for planning and Google Places/Unsplash for enrichment, and Sentry provides redacted error monitoring on Vercel.