🧾 InvoiceAudit - CrewAI Crew

📘 Description

InvoiceAuditCrew is an AI-powered crew that automates the extraction, categorization, and auditing of invoice documents. Using OCR and language models, it processes PDF invoices into structured formats, identifies policy violations, flags duplicates, and ensures all invoices are audit-ready. This reduces manual bookkeeping efforts and improves financial accuracy and compliance.

💼 Example Use Cases

• Freelancers and Consultants
  Automatically extract invoice data from scanned or emailed PDFs and generate categorized expense reports in seconds. Save hours each month otherwise spent on manual data entry, and ensure that income and expenses are ready for tax filing or client reimbursement.

• Accountants Handling Large Volumes
  Upload dozens (or hundreds) of client invoices and let the system extract line items, apply accounting categories (e.g., Software, Meals, Travel), and run compliance checks (e.g., tax rate validation, missing vendor info). Bulk-export CSVs compatible with QuickBooks or Xero.

• Small Business Owners Managing Vendor Payments
  Use the crew to catch duplicate or inflated charges across recurring invoices. For example, if a subscription charge appears twice with slightly different invoice numbers, the system will flag it. This prevents overpayments and improves accounts payable hygiene.

• Finance & Operations Teams in Growing Startups
  Integrate this crew into your internal invoice intake pipeline. Automatically route suspicious invoices (e.g., wrong currency, mismatched totals) to a human reviewer while clean ones go straight to approval. Great for teams scaling up without adding extra accounting headcount.

• Nonprofits & Grant-Funded Projects
  Generate audit-ready documentation for each invoice submitted to a grantor or funding agency. Quickly ensure all required fields (vendor name, date, description, currency) are present and properly categorized. Export a clean summary for reporting requirements.

• Auditors & Internal Compliance Teams
  Run automated policy checks across invoice datasets to ensure purchases comply with internal spending rules (e.g., maximum per diem meal costs, allowed vendors, proper tax inclusion). Easily identify potential violations ahead of formal audits.

📦 Dependencies

• crewai>=0.10.0
• pytesseract
• pdfplumber
• pdf2image
• pandas
• openai
• fuzzywuzzy
• dateparser

OCR Requirement:

• Tesseract OCR (must be installed separately)
  • Ubuntu: sudo apt install tesseract-ocr
  • macOS: brew install tesseract
  • Windows: Tesseract GitHub

API Keys Required:

• OpenAI API

---

⚙️ Configuration Instructions

1. Clone the repo

   git clone https://github.com/your-org/invoice_audit_crew.git
   cd invoice_audit_crew

2. Create a virtual environment

   python3 -m venv .venv
   source .venv/bin/activate

3. Install dependencies

   pip install -r requirements.txt

4. Set OpenAI API Key Create a .env file:

   OPENAI_API_KEY=your-key-here

5. Add invoice files Place your test invoices in input/invoice_sample.pdf or replace with your own.

---

🚀 Usage Examples

Run the full pipeline:

crewai run

Train the crew:

crewai train --n_iterations 3 --filename training_results.pkl

Test with different LLM models:

crewai test --n_iterations 2 --model gpt-4

---

🧾 Sample Output

📁 Output Directory: output/

• invoice_summary_YYYYMMDD_HHMM.csv: Categorized invoice lines with audit flags
• audit_report_YYYYMMDD_HHMM.pdf: Human-readable report of invoice findings

🧮 Example CSV:

A clean CSV file or QuickBooks-compatible report like:

Date,Vendor,Item Description,Amount,Currency,Category,Audit_Flag
2025-05-01,Adobe Inc.,Photoshop Subscription,29.99,USD,Software,
2025-05-03,Uber Eats,Team Lunch,120.00,USD,Meals,PolicyLimitExceeded
2025-05-04,Adobe Inc.,Photoshop Subscription,29.99,USD,Software,PossibleDuplicate

🧮 Output Example PDF

Below is a sample of the generated Audit Report PDF, which summarizes issues identified in each invoice in a human-readable format:

---

Audit Report – Generated: 2025-05-01

Invoice from: Adobe Inc.
Invoice Date: 2025-05-01
Item Description: Photoshop Subscription
Amount: 29.99 USD
Category: Software

Audit Results:

• ✅ No issues detected

---

Invoice from: Uber Eats
Invoice Date: 2025-05-03
Item Description: Team Lunch
Amount: 120.00 USD
Category: Meals

Audit Results:

• ❗ Policy Violation: Lunch expense exceeded allowed limit (PolicyLimitExceeded)

---

Invoice from: Adobe Inc.
Invoice Date: 2025-05-04
Item Description: Photoshop Subscription
Amount: 29.99 USD
Category: Software

Audit Results:

• ⚠️ Duplicate Detected: Similar invoice already exists

---

📁 This PDF is saved to: output/audit_report_YYYYMMDD_HHMM.pdf

📚 References

• Tesseract OCR
• CrewAI Framework
• QuickBooks CSV Guide

🤝 Contributing

Contributions are welcome! Whether it's new audit rules, integrations, or UI hooks, feel free to fork and submit a pull request.

---

📄 License

Licensed under the MIT License.