Architecture / Interactive guide

Understand the system.
One layer at a time.

Explore the role of each layer, the decisions within it and the connections that make it useful. This is a conceptual model, not a deployment blueprint.

CONCEPTUAL SYSTEMBring layers together
ApplicationAIDataFoundationSECURITY & OVERSIGHT
03 / INTELLIGENCE

The AI layer

Models use context and tools within the boundaries set by the system.

What matters here

  • Choose the model for the task
  • Evaluate the quality of its responses
  • Control tool access and consequential actions

Depends on relevant data. Supports the application.

Read the explanation Explore related work

Controls run
through every layer.

Security, evaluation and oversight belong throughout the system, not in a box added at the end.

01

Identity & permissions

Authenticate the user and authorise each retrieval or action. Keep permissions attached to the information.

02

Evidence & evaluation

Measure retrieval and answer quality separately. A fluent response is not proof that it is correct.

03

Operations & oversight

Monitor quality, cost and failure. Give people control over consequential actions and a route to recover.

absar.aiENTERPRISE AI / A VISUAL EXPLANATION
01 / THE QUESTION

A question starts
the journey.

An employee asks when an expense claim must be submitted. The system needs the relevant policy, not a plausible guess.

Illustrative scenario · Synthetic policy · No live AI

Explore the concept ↗
REQUESTWhen should I
submit my claim?
A real task, clearly framed.
IDENTITYEmployeeRole and access checked✓
EVIDENCEExpense policySubmit within 30 days.
Finance-only source Restricted
AI
Draft with contextRelevant evidence supplied
DRAFT ANSWERWithin 30 days
of completing the trip.
[1] Expense policy
EvidenceAccessRelevance
HUMAN REVIEWInspect. Decide. Act.Consequential actions stay controlled.
0:00 / 1:00
Read the complete explanation

A question starts the journey. Identify who is asking, then retrieve only sources they are authorised to use. Supply relevant evidence to the model and keep references attached to its draft. Check whether the answer is supported, relevant and appropriately scoped. Missing or conflicting evidence should be made visible. People retain control over consequential decisions. Infrastructure, security and monitoring support this whole workflow.