Platform · Architecture · US

Map AI Platform Architecture to Your Business Environment

A buyer-facing walkthrough of LuMay's six-layer reference architecture, aimed at the questions a US technical evaluator asks before approving a pilot: where does data flow, what are the system boundaries, and who owns each failure path?

Overview

LuMay's architecture is a six-layer reference model with defined contracts between layers: Interface, Intelligence, Orchestration, Integration, Governance, and Analytics & ROI. Each layer has a single responsibility and a defined set of inputs and outputs, which is what lets you change a model, replace a connector, or add a channel without touching the rest of the stack.

For a US evaluation, the practical question is rarely "is this architecture good" - it's "where does this sit relative to what we already have, and what has to change on our side." The sections below walk through that mapping.

Layered stack - platform spine highlighted
01InterfaceVoice, chat, Teams, web portal, REST API
02IntelligenceRAG, reasoning, classification, structured outputs
03OrchestrationWorkflows, multi-agent, HITL, versioning
04IntegrationConnectors, adapters, OAuth, MCP servers
05GovernanceRBAC, RLS, audit, PII controls, guardrails
06AnalyticsKPIs, cost, quality, traces, sentiment
Change one layer, preserve the rest
Model provider→Intelligence
New channel→Interface
Data migration→Integration
New policy→Governance
Deployment shift→Runtime shell

This is what 'platform, not project' means in practice.

The Six Layers

#LayerResponsibilityComponents
01InterfaceChannels agents are reached throughVoice (Twilio WebSocket), Chat, Microsoft Teams, Web portal, REST API
02IntelligenceReasoning and understandingRAG, LLM reasoning (OpenAI / Anthropic), intent classification, structured outputs
03OrchestrationWorkflow execution and multi-agent coordinationConversation flows, multi-agent delegation, HITL gates, versioning, NL builder
04IntegrationConnecting agents to external systemsConnectors (ticketing, email, docs), OAuth vault, MCP servers
05GovernanceTrust, compliance, and controlRBAC, Row-Level Security, audit logs, PII controls, guardrails, secrets management
06Analytics & ROIMeasurement and value demonstrationKPIs, cost tracking, ROI dashboard, sentiment, quality scores, OTel traces, Prometheus metrics

Mapping to Your Environment

Before scoping a pilot, most US evaluation teams work through this sequence against their own environment:

  • Interface (01) - which channels do you actually need on day one (voice, chat, Teams, API), and which systems currently own those channels today?
  • Intelligence (02) - which LLM provider(s) are already approved by your security team, and does your data governance policy allow the retrieval sources you plan to connect for RAG?
  • Orchestration (03) - which actions need a human approval gate before they execute, and who is that approver in your organization?
  • Integration (04) - which systems of record does the agent need read access to first, and which (if any) need write access?
  • Governance (05) - what is your existing identity provider, and does your tenant-isolation requirement match row-level security at the database layer, or do you need a harder boundary?
  • Analytics (06) - which business metric will you use to decide whether the pilot succeeded, and do you have a baseline for it today?

Illustrative Example

The following is an illustrative example of how a request might move through the six layers - it is not a specific customer deployment.

An internal assistant receives a request from an employee over chat (Interface). The request is classified and relevant context is retrieved via RAG (Intelligence). The orchestration layer determines the request requires an action in a ticketing system and checks whether that action is pre-approved or needs a human reviewer (Orchestration). If approved, the connector layer dispatches the action to the ticketing system using validated credentials (Integration), the action is logged with a correlation ID (Governance), and the outcome is recorded against the agent's containment-rate metric (Analytics).

Dependency Questions to Ask

  • If the LLM provider has an outage, which layer degrades, and does the rest of the stack keep functioning?
  • If a connector credential expires or a provider API changes, what is the failure path, and who is alerted?
  • If an approval gate is skipped by mistake, is that visible in the audit trail after the fact?
  • Who owns the decision to change a layer in production - engineering, security, or the business owner?

FAQ

Does changing LLM providers require re-architecting anything?

No - provider changes are contained to the Intelligence layer (02); agents, flows, and connectors are unaffected.

Can we deploy this entirely within our own cloud environment?

See Deployment for the specific models (SaaS, BYOC, private cloud, on-prem) and who manages infrastructure in each.

Where does tenant isolation get enforced?

At the Governance layer (05), via PostgreSQL Row-Level Security - see Security & Governance for detail.

What's the difference between this page and the business platform overview?

This page is the technical architecture reference. /platform covers the business case, pricing posture, and the four engines at a higher level.

Who should review this page before a pilot?

Typically a solutions architect or engineering lead, alongside whoever owns security/compliance sign-off for the systems being integrated.

Want this mapped to your systems of record?

Book a 30-minute architecture review. We'll walk through the six layers against your existing stack and flag anything that needs a custom connector or a deployment-model decision.