Skip to main content
Pedro Alizo

Finance Variance Review Workbench

  • Finance workflows
  • TypeScript
  • Synthetic data
  • Human review

Status: Interactive prototype

Open demo

Evidence highlights

  • Calculates variance amount, percentage, and direction, with explicit handling when the plan baseline is zero.
  • Lets a reviewer mark a synthetic exception reviewed or escalated; decisions remain in browser memory and are not written back.
Inspect source & tests (opens in new tab)

Project screenshots

Variance workbench queue showing plan-to-actual exceptions and review filters
Exception queue with filters and review state
Variance workbench detail panel with source references and review actions
Detail panel: sources, note, review / escalate

What a 30-minute review should inspect

Open the interactive demo and check:

  1. Zero-plan handling — variance percentage stays undefined instead of dividing by zero.
  2. Source references — synthetic identifiers sit beside the record they support.
  3. Review states — mark reviewed or escalated; counts and filters update; state stays in browser memory.
  4. Fixture honesty — context text is fixed fixture text, not generated reasoning.

Then skim the implementation source and tests for the calculation and review-state behavior.

Problem framing

Finance exception review requires more than a calculation. A reviewer needs to see the plan and actual, understand which source detail supports a proposed explanation, record a decision, and know what the tool did not verify.

This prototype makes that review path visible using invented operating-expense records and generic categories. It is a workflow demonstration, not an implementation for any employer or customer.

Status and evidence boundary

Status: interactive browser prototype. There are no LLM calls, file uploads, external requests, real finance data, connected source systems, persistent audit records, or accounting-system writebacks. No time savings, accuracy improvement, adoption, or production outcome is claimed.

A production deployment would require permission-aware integrations, finance-approved materiality rules, durable audit logging, security and privacy review, failure handling, and evaluation with domain-reviewed examples.