Alsvior AI Lab

Work you can inspect.

Demonstrations and templates for designing coordinated AI workflows. The onboarding scenario draws on patterns from real customer cases; every person, record, message and outcome shown uses sample data. The experiments are freely accessible and separate from client deployments.

Control room →Templates →Service economics →

Agent control room

ALSVIOR AI LAB / SIMULATION 01

Customer onboarding hive

Based on patterns from real customer cases. People, records, dialogue and outcomes here use sample data in a scripted simulation.

CASE-INSPIRED SIMULATION
No live customer data or systems

Static walkthrough: Maya receives an onboarding request, assigns customer success, identity, delivery and knowledge tasks, then asks Rae to test completion. A conflicting role source blocks access; a delivery timeout requires reconciliation before retry. Customer operations owns either exception. All records use sample data.

Change the case

CASE ONB-0240 / 24 EVENTSREADY

COORDINATION MAP00 KNOWLEDGE NODES

AgentKnowledge recordActive messageDrag nodes · drag background · scroll to zoom

HIVE / CASE CONVERSATION

PAUSED

Ready to inspect the work.

Play the case or advance one message at a time. Every exchange below is scripted from sample records.

SHARED KNOWLEDGE / LIVE CASE RECORD

Watch knowledge take shape.

Each node is created at a specific message. Select a record to see what it says, where it came from, who may use it and whether it is approved.

00records created

0 sources 0 candidates 0 approved / results 0 conflicts / exceptions

No records yet. Play the case to watch the first source appear.

No knowledge has been created yet.

Source records, candidate claims, validation results and exceptions will appear here as the conversation progresses.

All names, documents, permissions, dates and outcomes shown here are sample data. No source content or customer identity is exposed.

Agent roster

Presence follows assigned work

MayaOperations Lead Agentonline

Role in case: Orchestrator
Task: Sequence work and own the case
Permitted action: Assign, pause and route; no customer action
Inputs: Task contract and evidence
Result: Accepted outcome or owned exception

NoraCustomer Success Agentoffline

Role in case: Intake
Task: Validate the request
Permitted action: Read order and user list
Inputs: Order v3, user list v2
Result: Versioned intake record

AdaKnowledge Analyst Agentoffline

Role in case: Knowledge
Task: Check reusable facts
Permitted action: Propose or withhold scoped claim
Inputs: Source, scope, permission, expiry
Result: Approved or conflicted record

KaiIdentity Engineer Agentoffline

Role in case: Access
Task: Prepare access
Permitted action: Propose; no live grant
Inputs: Users and approved role claim
Result: Proposal and simulated test

LeoDelivery Coordinator Agentoffline

Role in case: Delivery
Task: Prepare handover
Permitted action: Draft; no external send
Inputs: Order, contact and scope
Result: Reviewed welcome draft

RaeQuality Assurance Agentoffline

Role in case: Verifier
Task: Test business completion
Permitted action: Accept or assign exception
Inputs: Branch outputs and evidence
Result: Case decision and evidence trail

Business outcome

STANDARD PATH

Pending acceptance

Agent outputs are intermediate. Completion requires evidence, acceptance tests and an accountable owner.

Keyboard: focus this panel and use ← / → for messages. Focus the graph and use arrow keys to pan, + / − to zoom, and Escape to reset. Drag nodes to stretch and release them; drag empty space to pan. Reduced-motion settings stop the ambient spring movement. All dialogue and records use sample data.

Use the operating templates.

Start with one repeatable job. Define its outcome and authority before adding agents. Adapt these examples to your own controls.

Task handover

Define owner, inputs, permitted action and completion test.

Download template

Knowledge record

Record provenance, scope, access policy, status and review date.

Download template

Completion contract

Set acceptance tests, authority, retry and exception owner.

Download template

Starter recipes

All four practical templates in one file.

Download template

Shared knowledge: promotion checklist

Capture the source, version and exact claim.
Check scope and source permissions; derived summaries inherit restrictions.
Compare candidate facts with current approved records and surface conflicts.
Assign an authorised reviewer or validation rule.
Set effective and review dates, then approve for the permitted use.
Mark superseded or expired facts unusable until reviewed.

AI service economics

Vary volume, failed attempts and review effort to see how an illustrative service changes.

Hypothetical monthly model. Edit the assumptions; no figures are client results.

Accepted jobsPrice per accepted job (£)Attempts per accepted jobCost per attempt (£)Jobs needing review (0–1)Minutes per reviewReviewer cost per hour (£)Fixed monthly costs (£)

Revenue: £20,000
All processing attempts: £1,725
Human review: £700
Operating remainder: £13,575

Formula: accepted jobs × price − all attempts − human review − fixed costs. Excludes setup investment, tax, financing, other variable delivery cost, working capital and cash collection timing.

From an experiment to a bounded pilot

The public examples have no live authority. A real pilot needs agreed permissions, test cases, acceptance criteria and an exception owner.

Explore the AI practice →