absar.ai

Labs / Guided experiments

Change an input.
Understand the difference.

Explore system behaviour through small, transparent experiments. Start with evidence and access control in a retrieval workflow.

Deterministic simulation

The evidence lab

A synthetic policy collection. No live model or external data.

01 / Set up your question

Role switching here is for learning. Real authorisation must be enforced on a trusted server.

02 / Inspect the result

Start with a question.

Run the simulation to see which source is eligible and how the response changes.

03 / Understand why

The simulation applies a role filter, checks simple question keywords and returns a predefined response with its source. It does not perform vector search or generate an answer with AI.

Inspect the synthetic source collection
SOURCE A · ALL EMPLOYEES

Expense submission policy

Submit eligible expense claims within 30 days of completing a trip.

SOURCE B · FINANCE REVIEWERS

Exception approval procedure

Exceptional expenses require approval from the finance director.

Invented policies for this demonstration only. Not business or financial guidance.

Connect the experiment to the architecture

Why does the evidence matter?

Read the explanation
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.