How LATAM Airlines Built Intelligent Agents in Aviation | Interrupt 2026

Short summary

LATAM Airlines shares how they built Concierge, a LangGraph-based travel agent serving 4,000 daily users in a high-constraint 3-5% margin industry. Key wins: 15% cost reduction through architecture redesign and discovery that 13% out-of-scope requests revealed product gaps, not model failures. They introduce Compass, an internal pipeline converting agent conversations into BigQuery knowledge graphs—shifting focus from agent-as-product to intelligence-as-product.

• •LATAM built Concierge agent on LangGraph serving 4K daily users in low-margin aviation
• •Achieved 15% cost reduction through architecture redesign; 13% out-of-scope rate was product gap, not model limitation
• •Compass pipeline transforms agent conversations into BigQuery knowledge graphs—the agent becomes infrastructure for intelligence extraction

Generated with AI, which can make mistakes.