Technology
The Menlo Tech Intelligence Engine
The Menlo Tech Intelligence Engine is a proposed, integrated architecture: five layers that turn enterprise data into governed, measured action. It is a reference design. Not every customer deployment contains every layer, and each component carries its maturity.
Maturity: published evidence means a public indicator or documented operation exists (Subway case); illustrative means a demonstration or tour; conceptual means a design not yet shown; roadmap means planned.
01 · Enterprise Context Layer
Purpose
Gather data from systems, spreadsheets, documents and email, resolve entities, and keep the source of every fact.
Components and maturity
- Data connectorsConceptual
- Entity resolutionIllustrative
- Knowledge graph with provenanceIllustrative
- Semantic index (embeddings)Illustrative
- Permission modelConceptual
Input
Raw company data and its sources.
Output
Structured context with origin and confidence.
Relationship to other layers
Feeds the Reasoning Engine and the agents; receives access rules from Execution & Control.
Public evidence
The Subway case describes titles arriving from an ERP, spreadsheets and email into a structured portfolio (explanatory model).
Appears in
02 · Reasoning Engine
Purpose
Interpret language and documents, estimate probabilities and choose actions with explicit mathematics.
Components and maturity
- Language models for interpretationIllustrative
- Intent and document classificationIllustrative
- Confidence calibrationIllustrative
- Optimization and decision theoryIllustrative
- Forecasting modelsConceptual
- Model routingRoadmap
Input
Structured context and objectives.
Output
Ranked decisions with rationale and trade-offs.
Relationship to other layers
Consumes context; hands choices to Agent Orchestration; receives calibration from Evaluation & Learning.
Public evidence
Semantic retrieval and intent classification appear in the RE/MAX tour; confidence thresholds in the Clave tour. Both are illustrative.
Appears in
RE/MAX · Clave · Subway Brasil
03 · Agent Orchestration
Purpose
Plan and coordinate specialized agents, keep state and memory, and call tools.
Components and maturity
- PlannerConceptual
- Specialized agents (communication, management)Published evidence
- Specialized agents (others)Illustrative
- State machinesIllustrative
- Tool integrations (MCP)Conceptual
- Long-term memoryRoadmap
Input
Choices from the Reasoning Engine, plus policy.
Output
Proposed actions for each agent.
Relationship to other layers
Every proposed action passes through Execution & Control before it runs.
Public evidence
Subway: a communication agent (WhatsApp and email, 24/7, first response within one minute in 100% of cases) and a management agent are part of the published operation.
Appears in
RE/MAX · Clave · Subway Brasil
04 · Execution & Control
Purpose
Validate each action against policy, execute it, escalate exceptions to people and record every event.
Components and maturity
- Policy guardsPublished evidence
- Communication channelsPublished evidence
- Human approval queueIllustrative
- Event logIllustrative
- Idempotent, parallel executionConceptual
Input
Proposed actions and policy.
Output
Executed actions, escalations and events.
Relationship to other layers
Feeds Evaluation & Learning with events; returns access rules to the context layer.
Public evidence
Subway: the network defines the policy (criteria, control points and how far to go with a late franchisee) and Menlo Tech executes within those limits.
Appears in
05 · Evaluation & Learning
Purpose
Measure outcomes, monitor calibration and drift, and recalibrate weights and thresholds from evidence.
Components and maturity
- Outcome indicatorsPublished evidence
- Evaluation pipelinesConceptual
- Calibration and drift monitorsRoadmap
- Human-feedback loopRoadmap
- Threshold and weight recalibrationRoadmap
Input
Events, decisions and outcomes.
Output
Metrics, alerts and updated parameters.
Relationship to other layers
Closes the loop: recalibrates the Reasoning Engine and refines policy with the people who own it.
Public evidence
Only the Subway outcome indicators are published. The monitors and recalibration are a proposal.
Appears in
Maturity of the components
Of the 27 components in the reference architecture, this is what each level means today.
"Published evidence" is limited to what the Subway case documents publicly. Everything else is illustrative, conceptual or planned.
A synthetic request through the layers
Synthetic trace for illustration. No real customer data and no message is sent.
- +0 msContextTitle T-0421 arrives from the ERP and from a spreadsheet
- +38 msContextEntity resolved: both rows refer to Store 0421
- +61 msReasoningPriority score computed; payment probability estimated
- +74 msOrchestrationNext action proposed: light reminder by WhatsApp
- +80 msControlPolicy guard: amount and attempts within limits, so execute
- +95 msControlMessage dispatched (simulated)
- +96 msEvaluationEvent recorded; outcome pending