Multi-Agent AI System for Document Processing (Groq Llama3-8B Version)

This system accepts input in PDF, JSON, or Email (text) format, classifies its format and intent, and routes it to the appropriate specialized agent for processing. It maintains a shared context (memory) for traceability. This version is specifically configured to use the Groq API with the llama3-8b-8192 model.

System Overview

1. Input:
   • Raw file (PDF, JSON, TXT representing email).
   • Raw text string (simulating email content).
   • Raw JSON string.
2. Classifier Agent:
   • Determines the input format (PDF, JSON, Email/Text).
   • Extracts text content (for PDF/Text).
   • Uses the Groq Llama3-8B model to classify the intent (e.g., Invoice, RFQ, Complaint, Regulation) and urgency.
   • Logs format, intent, source, and other details to shared memory.
   • Routes the input (or extracted text/parsed JSON) to a specialized agent.
3. JSON Agent:
   • Accepts structured JSON payloads.
   • If the classified intent is "Invoice" (or as configured), it validates against a target schema (JSON_TARGET_SCHEMA in config.py).
   • Extracts/reformats data based on the schema.
   • Flags anomalies or missing fields.
   • If the intent is not "Invoice" or doesn't match schema criteria, it processes the JSON as a generic payload.
   • Logs results to shared memory.
4. Email Agent:
   • Accepts email content (text, including text extracted from PDFs).
   • Uses the Groq Llama3-8B model to extract key information (sender, refined subject, refined summary, urgency, action items, contact person/phone).
   • Formats the extracted information for CRM-style usage according to EMAIL_CRM_FORMAT_KEYS in config.py.
   • Logs results to shared memory.
5. Shared Memory Module:
   • A lightweight in-memory store (Python dictionary).
   • Stores: conversation ID, source, format type, timestamp, classified intent, extracted values from agents, errors, and processing steps.
   • Accessible across all agents for context and traceability.

Tech Stack

• Python 3.9+
• Groq API with the llama3-8b-8192 model for all LLM tasks.
• pdfplumber for PDF text extraction.
• Standard Python json library.
• In-memory Python dictionary for Shared Memory (memory/shared_memory.py).

Folder Structure

├── agents/ # Agent implementations (classifier, json, email)
│ ├── init.py
│ ├── base_agent.py
│ ├── classifier_agent.py
│ ├── json_agent.py
│ └── email_agent.py
├── memory/ # Shared memory module
│ ├── init.py
│ └── shared_memory.py
├── utils/ # Utility functions (LLM interaction, PDF parsing)
│ ├── init.py
│ ├── llm_utils.py
│ └── pdf_parser.py
├── sample_inputs/ # Example input files (JSON, TXT, and a placeholder for PDF)
│ ├── complaint_sample.txt
│ ├── invoice_sample.json
│ ├── regulation_sample.txt
│ └── rfq_sample_text_fallback.txt
├── main.py # Main script to run examples and orchestrate agents
├── config.py # Configuration (API keys, model names, schemas)
├── requirements.txt # Python dependencies
├── README.md # This file
└── .env # For API keys (gitignored)

Setup

1. Clone the repository (if you have one):

   git clone <your-repo-url>
   cd multi_agent_system

   (If you don't have a repo, just ensure all files are in the correct structure shown above.)

2. Create a virtual environment (recommended):

   python -m venv venv
   source venv/bin/activate  # On Linux/macOS
   # venv\Scripts\activate    # On Windows

3. Install dependencies:

   pip install -r requirements.txt

4. Set up Groq API Key:

   • Sign up at GroqCloud to get an API key if you don't have one.
   • Create a file named .env in the root of the multi_agent_system directory.
   • Add your Groq API key to it:

     GROQ_API_KEY="your_groq_api_key_here"
     

How to Run

Execute the main.py script from the root of the multi_agent_system directory:

python main.py