When companies fly out candidates for interviews, they often give candidates the ability to reimburse their travel and meals. Unlike regular employees, these candidates are not in the company's expense management system. Instead, they are often told to take pictures of their receipts and email them to HR, where managers have to manually review each receipt to ensure compliance with budgets and policies.

To automate this process, we built a pipeline using the Vector Shift platform. The pipeline works as follows:

1. An internal email is set up for the company, where candidates can forward their receipt emails.
2. The pipeline extracts data from the receipts, including the category of spend (food, travel, or other), the total amount, and the vendor name.
3. The pipeline categorizes the spend and checks if the total spend per category is within the company's approved limits (e.g., $50 for food, $100 for travel).
4. The pipeline sends an email back to the HR manager, indicating whether the spend is approved and the total reimbursable amount.
5. The pipeline also records the reimbursement details in a Google Sheet for record-keeping.

By automating this process, the company can streamline the receipt reimbursement for candidates, reducing the manual effort required by HR and ensuring compliance with company policies.

To automate the receipt reimbursement process, we first need to categorize the receipts and extract the key data from them. This is done using a combination of machine learning and data extraction techniques:

1. Categorization: We use a categorizer node powered by a large language model (LLM) to classify the receipt into one of three categories: food, travel, or other. The categorizer is trained on example text for each category to help it accurately identify the type of expense.

2. Data Extraction: We then extract the total amount spent and the vendor name from the receipt text using another set of LLM-powered extraction nodes. These nodes are provided with descriptions of what to look for in the receipt text to identify the relevant data points.

3. Output: The categorized spend type, total amount, and vendor name are then output from the pipeline, ready to be used in the next steps of the reimbursement process.

By breaking down the receipt processing into these discrete steps, we can leverage the strengths of LLMs to accurately handle the variability in receipt formats and content, while maintaining a structured and reliable pipeline.

To calculate the total spend by category, we first extract the category and amount data from the receipts using the "Process Individual Receipt" pipeline. This pipeline categorizes each receipt as either "food", "travel", or "other", and extracts the total amount and vendor name.

The extracted category and amount data is then passed to an OpenAI node, which creates a summary describing the total spend per category. This summary is then evaluated by another OpenAI node to determine if the total spend is within the approved limits set by the HR department (e.g., $50 for food, $100 for travel).

Based on the evaluation, the pipeline generates an email response to the sender, indicating whether the total spend is approved and the reimbursable amount. Additionally, the pipeline records the details (email, vendor, category, and spend) in a Google Sheet for record-keeping purposes.

This multi-step process, with each step handled by a separate OpenAI node, helps improve the accuracy and reliability of the receipt reimbursement pipeline by breaking down the task into more manageable components.

The pipeline we've built automates the process of approving candidate travel and meal reimbursements within a set budget. Here's how it works:

1. The pipeline starts when a new email is received in the designated inbox, containing receipt attachments from candidates.
2. The pipeline processes each receipt attachment individually, using a categorizer node to classify the expense as either food, travel, or other. It also extracts the vendor name and total amount spent.
3. The pipeline then aggregates the total spend per category and compares it against the approved limits (e.g., $50 for food, $100 for travel).
4. Based on the summary of expenses, the pipeline generates an email response to the sender, indicating whether the total spend is approved and the reimbursable amount.
5. Finally, the pipeline logs the expense details (email, vendor, category, and amount) in a Google Sheet for record-keeping.

This automated process saves HR teams significant time and effort, allowing them to efficiently manage candidate reimbursements while ensuring compliance with company policies.

To ensure a comprehensive record of the reimbursement process, the pipeline integrates with a Google Sheet to track and store the relevant data. The following steps are taken:

1. **Create a Google Sheet**: A Google Sheet named "receipt reimbursements" is created to store the reimbursement data.

2. **Define the Sheet Columns**: The sheet includes the following columns:
   - Email: The email address of the person submitting the reimbursement request.
   - Vendor: The vendor name extracted from the receipt.
   - Category: The category of the expense (food, travel, or other).
   - Spend: The total amount spent, as extracted from the receipt.

3. **Populate the Sheet**: The pipeline uses the "Column List Writer" node to write the data extracted from the receipts into the corresponding columns of the Google Sheet. This includes:
   - Vendor name
   - Expense category
   - Total spend amount

4. **Handle Multiple Receipts**: If there are multiple receipts attached to the email, the pipeline uses the "Duplicate" node to create a list of the sender email addresses, matching the number of receipts. This ensures that each row in the Google Sheet is associated with the correct email address.

5. **Automate the Process**: The pipeline is triggered by new emails received in the designated inbox. This allows the reimbursement process to be automated, with the pipeline automatically extracting the data, categorizing the expenses, and updating the Google Sheet with the relevant information.

By integrating with a Google Sheet, the pipeline ensures that a comprehensive record of the reimbursement process is maintained, providing visibility and traceability for the HR team.

Here is the section body in Markdown format:

In this video, we have demonstrated how to build a receipt reimbursement pipeline using the Vector Shift platform. This pipeline automates the process of managing travel and expense reimbursements for job candidates, which can be a time-consuming and manual task for HR departments.

The key steps involved in this pipeline are:

1. Extracting data from the receipt, including the category of spend (food, travel, or other), the total amount, and the vendor name.
2. Categorizing the spend based on the extracted data and comparing the total spend in each category against the company's reimbursement limits.
3. Generating an email response to the HR manager, informing them of the approved reimbursement amount.
4. Logging the reimbursement details in a Google Sheet for record-keeping.

By automating this process, companies can streamline their candidate reimbursement workflow, reducing the administrative burden on HR teams and ensuring consistent and compliant reimbursement decisions.