Expense Reimbursement Agent

Agent Description:

The Expense Reimbursement Agent is an AI-powered system designed to automate the end-to-end process of employee expense claim verification. It extracts data from digital receipts, validates claims against company policy, and summarizes reimbursement decisions for both employees and the finance department.

Purpose and Components
  • Purpose: To automate the validation and processing of expense claims from digital receipts using OCR, policy checks, and summary generation.
  • Components:
    • A receipt extractor to digitize and structure raw receipt content.
    • A validator assesses policy compliance for each item.
    • A final summarizer to communicate approval status.
Supported Capabilities
  • Extracting structured data from PDFs and image-based receipts.
  • Standardizing receipt data into consistent formats (for example, date, currency).
  • Checking claims against reimbursement policies using a knowledge base.
  • Categorizing each claim line as Approved, Rejected, or Flagged for Review.
  • Generating final reports for finance teams and employees.
LLM Used
  • Google Vertex
    Note: To learn more about the LLM and to modify its behavior, refer to the Configuring LLM settings section.

Sub-Agents

1. Receipt Extractor Agent

  • Role:Receipt Extractor.
  • Scope:Process submitted files to pull vendor name, date, items, and amounts.
  • Description:

    • Accept receipt input (for example, uber_receipt.jpg, hotel_invoice.pdf).

    • Use OCR to convert to raw text.

    • Extract fields:

      • Vendor Name
      • Transaction Date
      • Itemized Items (with price)
      • Total Amount
      • Currency

    • Correct common OCR errors based on text context.

    • Normalize:

      • Dates → YYYY-MM-DD

      • Amounts → Float (for example, ₹1450.50)

  • LLM Used: Google Vertex (inherits from parent).
  • Tools used: PDF Reader.

2. Expense Validator Agent

  • Role:Expense Validator.
  • Scope:Validate expenses using a RAG-based search on internal reimbursement policy.
  • Description:

    • Input: structured receipt JSON from Receipt Extractor.

    • For each item:

      • Use RAG over company_expense_policy.txt to verify if reimbursable.

      • Evaluate cost against limits, exclusions (for example, alcohol, international roaming).

    • Categorize as:
      • Approved
      • Rejected
      • Flagged for Review
  • LLM Used: Google Vertex (inherits from parent).
  • Tools used: MCP Client Tool

3. Expense Reimbursement Agent

  • Role:Expense Summarizer.
  • Scope:Create a reimbursement summary report for employee and finance teams.
  • Description:
    • Input: Validated receipt data from Validator Agent.
    • Group by status:
      • Approved
      • Rejected (with reason)
      • Pending Review
    • Compute totals:
      • Total Approved
      • Total Rejected
      • Total Pending
  • LLM Used: Google Vertex (inherits from parent).
Tools Used:

MCP Client Tool: for structured file handling and routing

Note: For details on modifying the Tools, refer Tools Library section.
Agent Workflow Behavior Summary
  • The Receipt Extractor Agent digitizes and parses receipt content from uploaded files.
  • The Expense Validator Agent checks every item against the company policy using a retrieval-based search and flags any issues.
  • The Expense Reimbursement Agent compiles the outcome into a concise, structured summary for final submission to finance or for employee feedback.
Sample Questions:
  • Can you check if this hotel receipt is reimbursable under our travel policy?
  • Does this receipt violate any reimbursement rules