📟 SmartAudit LLM: Legal + Managerial + Financial Agent

🔍 Overview

SmartAudit LLM is an advanced, autonomous auditing platform designed to analyze and validate financial documents such as invoices in CSV and PDF formats. The system leverages a streamlined architecture that relies entirely on Llama 3 via Groq to handle both deep reasoning and rule-based validation, creating a cohesive and powerful auditing engine.

• Unified LLM Intelligence: Uses Llama 3 (via Groq) for all tasks, including pattern detection, compliance checks, and generating natural language summaries.
• Rule-based Logic: Implements strict compliance and business rules to catch errors, fraud, and inconsistencies.
• Multi-Agent Design: Simulates three expert roles—Legal, Managerial, and Accountant—each providing specialized checks and insights.

What It Does

1. File Upload & Parsing: Users upload invoice files (CSV or PDF) through a modern React frontend. The backend parses and normalizes the data into a structured format.

2. Automated Auditing: The FastAPI backend orchestrates the audit process. It applies rule-based checks (powered by Llama 3) to detect:

   • Missing or invalid legal fields (e.g., GSTIN, invoice ID)
   • Future-dated invoices
   • Vendor concentration and duplicate patterns
   • Calculation mismatches (e.g., total ≠ quantity × unit price)
   • Zero or negative values in numeric fields

3. Role-Based Analysis: The audit results are processed by Llama 3 to generate specialized insights:

   • Legal Summary: Compliance issues, missing fields, and legal risks
   • Manager Summary: Vendor risks, operational patterns, and managerial suggestions
   • Accountant Summary: Numeric consistency, calculation errors, and accounting recommendations

4. Visual Reporting: The frontend displays the results as role-based markdown summaries, interactive tables, and visual dashboards. Users can export the findings as PDF or markdown reports.

5. Extensible & Modular: The architecture is agentic and modular, making it easy to add new rules, models, or UI features.

---

SmartAudit LLM combines the power of state-of-the-art LLMs and traditional audit logic to deliver a comprehensive, explainable, and user-friendly financial document auditing solution.

---

🏗️ System Architecture

SmartAudit LLM is an autonomous, multi-agent system for intelligent auditing of financial documents (CSV/PDF). It combines legal, managerial, and financial expertise using a unified Llama 3 architecture. The system features:

• Llama 3 via Groq (Generative): Generates markdown summaries and role-based insights (Legal, Manager, Accountant).
• Llama 3 via Groq (Logic): Performs rule-based audit logic (duplicate detection, total mismatch, compliance checks).
• React Frontend: User interface for uploading files, visualizing audit results, and exporting reports.
• FastAPI Backend: Handles file processing, audit orchestration, and API endpoints.

---

🚀 Features

• Upload and audit invoices in CSV or PDF format
• Automated legal, managerial, and accounting checks
• Role-based markdown summaries for easy understanding
• Visual dashboards and exportable reports
• Modular, agentic architecture for extensibility

---

📂 Project Structure

Frontend/               # React-based frontend UI
Backend/                # FastAPI backend for coordination
parsers/                # File input and parsing logic
├── csv_parser.py
└── pdf_parser.py
sample_data/            # Sample CSV and PDF invoices for testing
llama/                  # Summary generation and prompt logic
misrtal/                # Audit tools and rule-based logic (Llama 3 just file name)
README.md               # Project documentation
requirements.txt        # Python dependencies
test.py                 # Python test file

---

🧑‍💼 Agent Roles & Logic

⚖️ Legal Agent

• GST compliance checks (GST%, GSTIN format)
• Flags future invoice dates
• Identifies missing legal fields (vendor, invoice ID)

👔 Manager Agent

• Detects vendor concentration and high-value invoices
• Flags frequent identical amounts and repeated patterns
• Aggregates total spending per vendor

🧮 Accountant Agent

• Detects total mismatches (total ≠ quantity × unit_price)
• Flags missing/invalid numeric fields (zero/negative values)
• Ensures numerical consistency across invoices

---

🔄 Workflow

graph TD
    A["User Uploads Invoice File (CSV/PDF) - React UI"] --> B["FastAPI Backend (Parse & Normalize)"]
    B --> C["Llama 3 Audit (Validation & Compliance Agent)"]
    C --> D["Audit JSON Output"]
    D --> E["Llama 3 Prompting (Summary, Roles, Suggestions)"]
    E --> F["Markdown / PDF Report"]
    F --> G["React UI (Render Cards, Tables, Charts)"]

Loading

---

📦 Installation

1. Clone the repository

2. Install Python dependencies:

   pip install -r requirements.txt

3. Install frontend dependencies:

   cd Frontend
   npm install

4. Set up environment variables (see .env.example if provided)

---

🧪 Example JSON Input (to Audit Engine)

[
  {
    "invoice_id": "INV-XXXX",
    "date": "YYYY-MM-DD",
    "vendor": "Vendor Name",
    "products": [
      {
        "name": "Item Name",
        "quantity": 0.0,
        "unit_price": 0.0,
        "total": 0.0
      }
    ]
  }
]

---

🧾 Example JSON Output (from Audit Engine)

{
  "summary": {
    "total_invoices": 0,
    "vendors": 0,
    "date_range": {
      "start": "YYYY-MM-DD",
      "end": "YYYY-MM-DD"
    }
  },
  "issues": [
    {
      "invoice_id": "string",
      "vendor": "string",
      "issue_type": "string",
      "description": "string",
      "severity": "low | medium | high"
    }
  ],
  "compliance_flags": {
    "future_dates": [ { "invoice_id": "string", "date": "YYYY-MM-DD" } ],
    "missing_fields": [ { "invoice_id": "string", "field": "string" } ],
    "invalid_gstin": [ { "invoice_id": "string", "gstin": "string" } ]
  },
  "vendor_summary": [
    {
      "vendor": "string",
      "invoice_count": 0,
      "total_billed": 0.0
    }
  ],
  "invoice_patterns": {
    "duplicate_amounts": [ { "amount": 0.0, "invoice_ids": ["string"] } ],
    "repeated_items": [ { "item": "string", "occurrences": 0 } ]
  }
}

---

👥 Authors

• Sourish – Llama 3 + Summary Generator, Backend
• Prem – Audit Logic & Rule-based Tools
• Parth – Frontend (React), UI/UX