Step 1 · Connect estate

Two payroll systems, one wage definition

WageLens reads each system's pay-component dictionary and your policy documents. Salary rows stay where they are until the deterministic engine runs.

Source documents

Probabilistic planeAgent reads metadata

Component names, codes, descriptions and documents. About one model call per pay component.

Human gatePayroll head approves glossary

Every label carries a quote verified against the source text. Overrides need a reason.

Deterministic planeEngine computes every row

Each employee-month gets a trace with formula and clause. Re-runs cost nothing.

Step 2 · Agent run

Discovery agent

Profiles both dictionaries, merges duplicate components across systems, retrieves the law and policy, and proposes a clause for each component. Every output passes schema and quote checks before a person sees it.

Steps0

Tool calls0

Tokens0

Model time0 ms

Cost (est.)$0.000

Run log

Recorded run

Component resolution

Guardrail events

Classify a new pay component live

Step 3 · Review glossary

Approve one label per component

Sorted by risk: low confidence and large ₹ swing first. The ₹ figure is the change in annual PF plus gratuity liability if the label flipped, recomputed by the engine over the estate.

jk movea approveo overrideEsc close

Step 4 · Results

Liability impact

Deterministic engine over every employee-month in the estate.

Step 5 · Rules & tests

Versioned rules, hand-checked tests

Law changes are a ruleset version, not a code change. Switch versions and the engine shows exactly which employees move and by how much.