Skip to content
Kml
All projects
Company Project

AP Exception Handling Agent

B2B Accounts Payable automation solution that helps identify and handle invoice exceptions through AI-assisted workflows.

  • React
  • TypeScript
  • FastAPI
  • UiPath
AP Exception Handling Agent home screen on a desktop monitor, showing the suspended-balance summary, exception counts by status, a top-exceptions table, and the four-step ingest, validate, approve and post workflow.

Overview

An Accounts Payable automation product built at Auxiliobits, presented to users as LedgerClear. Invoices land in an exception queue, where an AI agent classifies and checks each one, routes it through configurable rules, drafts notes for the vendor or controller, and keeps a complete audit trail so finance teams can approve with confidence.

The Problem

Suspended AP cash builds up when invoices get stuck as exceptions: approvals that are still pending, vendor details that need validation, possible duplicates, and cases that need manual review. Investigating each one by hand is slow, and the reasoning behind decisions is hard to trace afterwards.

Architecture

  1. Invoice sources

    • Google Drive
    • SharePoint
    • Zoho

  2. Processing

    • AI agent: classify & check
    • UiPath automation
    • FastAPI services

  3. Review

    • Clearing desk (React + TypeScript)
    • Approvals in Slack

  4. Outcome

    • Posted to BILL.com
    • Complete audit trail
Vendor records and configurable rules drive validation and routing.

How It Works

  1. 1

    Ingest & classify

    Invoices from Google Drive, SharePoint and Zoho are captured and classified by the AI agent.

  2. 2

    Validate & route

    Each invoice is checked against vendor records and rules, then routed for approval.

  3. 3

    Approve in Slack

    Approvers get a clear view of each exception, with one-click decisions and notes.

  4. 4

    Post to BILL.com

    Approved invoices are posted to BILL.com with a full audit trail.

My Contribution

  • Developed Accounts Payable exception-handling automation.
  • Built backend services with Python and FastAPI.
  • Worked on UiPath workflows and their integration with the application layer.
  • Contributed to the React + TypeScript frontend, testing, debugging, and deployment.

Engineering Challenges

  • Handling distinct exception types (pending approval, vendor validation, duplicates, manual review) with clear routing for each.
  • Connecting several external systems, from document sources to Slack and BILL.com, into one dependable workflow.
  • Keeping AI-assisted decisions reviewable and fully audited, so people stay in control of financial actions.

Lessons Learned

  • Automation delivers the most value when the exception path is designed as carefully as the happy path.
  • Meeting approvers where they already work, such as Slack, removes friction from the approval step.