The AI step that knows its place
D4's LMM module is architecturally isolated. It receives curated input, returns structured output, and cannot touch anything else in the platform.
Where it sits
Architecturally isolated, by design
The LMM is not embedded in D4's core stack. It connects as a 3rd party module via API, the same way GEO lookup and weather services do. D4 controls what goes in and what comes out.
d4 orchestration
Filtered stream
Structured, verified data from streams layer.
Prompt handling
D4 defines the question. Controls exactly what the LMM receives.
Analyse engine
D4 processes the LMM's response before it becomes a result.
Enriched result
Versioned output, traceable and auditable.
3rd party
Replaceable
alongside
d4 orchestration
Filtered stream
Structured, verified data from streams layer.
Prompt handling
D4 defines the question. Controls exactly what the LMM receives.
Analyse engine
D4 processes the LMM's response before it becomes a result.
Enriched result
Versioned output, traceable and auditable.
3rd party
Replaceable
alongside
Constraints
What the AI module cannot control
These constraints aren't gaps in capability. They are architectural boundaries that let AI contribute to workflows without giving it control of the system.
No access to raw data streams
The LMM receives curated, structured input prepared by D4. Raw sensor data, telemetry feeds, and unfiltered events never reach the model directly.
No direct write access
The LMM can return responses that inform what happens next, but it cannot directly modify records or data in the platform. Any action must pass through the defined workflow.
Not the decision-maker
The LMM produces structured insight that can inform or trigger the next step in a workflow. D4's orchestration layer determines how that output is handled, validated, and acted upon.
Cannot affect upstream data
The LMM operates within the orchestration layer. It receives data from defined inputs and cannot modify or influence the data that was processed earlier in the flow.
The controls
D4 controls every boundary
Three enforcement points surround the AI step. Together, they keep every AI interaction scoped, validated, and traceable.
The prompt
D4 defines exactly what the LMM receives. The input is a curated summary constructed from filtered stream data, not a raw feed. The question, context, and response format are defined by the workflow.
The output
D4 receives the LMM's response and validates it before it becomes a result. Responses that don't match the expected structure or constraints are rejected or handled by the workflow.
The audit trail
AI interactions are versioned and timestamped. The input, response, and resulting output can be logged with the context needed to trace how a result was produced.
The governed AI step is one part of the full platform
It sits inside the orchestration layer, alongside GEO lookup, weather feeds, and your custom business logic. Book a demo and we'll show you the full chain, applied to your data.
