RAG AGENT USING CREWAI

🔍 What is this project?

This project is a RAG Agent (Retrieval-Augmented Generation Agent) built with CrewAI. It demonstrates how a RAG agent can answer queries asked by grounding its responses in a knowledge base (PDF), here in this case "RAG_MAS.pdf" -> "RAG-Multi Agent System"

Instead of relying only on the LLM, the RAG Agent uses:

• 🔎 Retrieve → Find relevant context from PDFs
• 🧠 Augment → Combine retrieved context with user query
• ✍️ Generate → Produce grounded, factual responses

👉 In simple words: This is a RAG Agent that answers only with facts from your PDF, not hallucinations.

⚙️ Tech Stack

• 🤖 CrewAI – Agent framework
• ⚡ Groq (ChatGroq) – LLaMA 3.1 8B Instant model
• 🧩 HuggingFace – all-MiniLM-L6-v2 embeddings for semantic search
• 📚 RagTool (CrewAI) – Retrieval-Augmented tool for document Q&A
• 🐍 Python – Core implementation
• 🔑 dotenv – For API key management
• 🏦 Chromadb - Since we didn't mention vectordb specifically, CrewAI defaults to Chroma.

🏗️ System Design:

The project defines RAG Agent:

1. 📄 SOW RAG Agent (sow_agent)
   • Role: Senior SOW Assistant
   • Goal: Help answer questions about Statements of Work
   • Backstory: Expert in contracts and compliance, designed to retrieve facts from documents
   • Tools: RAG Tool for PDF-based retrieval

📂 Project Workflow:

1. Load your PDF knowledge base (e.g., Rapid SOW Generation Guide).

2. Initialize the RagTool with this source.

3. Define the SOW RAG Agent (sow_agent) with:

   • Role, Goal, Backstory
   • Groq LLaMA 3.1 8B Instant as its reasoning model
   • HuggingFace MiniLM embeddings for retrieval

4. Create Tasks that query the RAG agent, for example:

   • “What is the waiting period for rehabilitation?”
   • "What is the standard timeline for completing a Statement of Work draft?"
   • "Before final approval, what review steps must the SOW undergo?"

5. Run the task → The RAG agent retrieves context from the PDF and generates a grounded answer.

🎯 Why this project matters:

• Shows how to build a RAG Agent with CrewAI
• Demonstrates Groq LLMs (LLaMA 3.1 8B Instant) in action
• Uses HuggingFace embeddings for document retrieval
• Grounded in a real-world PDF knowledge base

📌 Key Takeaways:

• This is not a chatbot – it’s a RAG Agent
• RAG (Retrieval-Augmented Generation) = making LLMs reliable by grounding them in documents
• Agent = AI worker with a defined role (here: SOW Assistant)
• CrewAI = orchestrates the RAG Agent + tools