Section 12
One sentence, seven layers, 1.2 seconds.
Press play and watch a real request travel the stack. Every layer below is owned by VOYA — source, infrastructure and data — with partners isolated behind a single adapter boundary.
The intelligence layer. Intent classification, conversation state, tool selection, retrieval over memory and content, and the guardrails that stop the agent acting beyond its authority.
Intent & entity extractionTool registry & executorConversation state storeMemory retrieval (vector)Prompt & version managementGuardrails & PII redactionModel routing & fallbackCost & latency telemetry
Request trace
Non-functional targets
Chat first token< 800 ms
Search p95< 2.5 s
Booking p95< 6 s
Availability99.9% monthly
RPO / RTO15 min / 1 h
Concurrent usersDesigned for 50k
Scaling path
Beta runs comfortably on a modest footprint. Search, orchestration and adapters scale horizontally and independently, so a traffic spike on hotel search never slows the chat experience or the booking saga.