🧠 Framework-Free RAG: Build a Retrieval-Augmented Generation App from Scratch

By Jay Patel — Software Developer | AI Enthusiast

This project demonstrates how to build a fully functioning Retrieval-Augmented Generation (RAG) application without using LangChain or LlamaIndex. It leverages:

• 🧾 Local documents (.pdf, .docx, .txt)
• 💬 OpenAI API (GPT-4 or GPT-3.5)
• 🔍 ChromaDB as a vector store
• 🧠 Conversational memory for multi-turn dialogue

---

📁 Project Structure

rag-from-scratch/
│
├── docs/                # Sample documents (.pdf, .docx, .txt)
├── main.py             # One-file full RAG implementation
├── requirements.txt    # Python dependencies
└── README.md          # You're here!

---

📦 Installation

1. Clone the Repository

git clone https://github.com/pateljay9936/RAG_Without_Framework.git
cd RAG_Without_Framework

2. Install Dependencies

pip install -r requirements.txt

---

📄 Supported File Formats

This app processes:

• .txt → plain text summaries
• .pdf → scanned reports and whitepapers
• .docx → internal documentation and company histories

---

🚀 How to Use

1. Prepare Documents

Add your .txt, .pdf, and .docx files into the docs/ directory.

2. Run the Application

python main.py

3. Ask Questions!

You'll be prompted to enter a query like:

> When was GreenGrow Innovations founded?

And get a response like:

Answer: GreenGrow Innovations was founded in 2010.
Sources: GreenGrow Innovations_Company History.docx (chunk 0)

---

⚙️ Key Features

✅ Document ingestion from local files
✂️ Smart text chunking (~500 characters)
🔍 Semantic search with sentence-transformers + ChromaDB
💬 GPT-4 or GPT-3.5 responses via OpenAI API
🧠 Conversational memory to handle follow-ups

---

🧪 Sample Queries

• Q: When was GreenGrow Innovations founded?
• Q: Where is it headquartered?
• Q: What is EcoHarvest?
• Q: Which companies focus on biotech research?

---

🧠 Powered By

• ChromaDB
• Sentence Transformers
• OpenAI API
• Python 3.8+

---

🧰 Want to Extend It?

You can plug in:

• Your company's internal documentation
• Custom OpenAI prompts
• LangGraph/agentic memory if you want to scale up later

---

📬 Contact

Made with ❤️ by Jay Patel

---

Let me know if you'd like this in a downloadable .md file or published to GitHub!