ChatPDF Application: Step-by-Step Installation Guide

ChatPDF is a locally hosted application that enables users to interactively query the contents of PDF documents. It leverages Retrieval-Augmented Generation (RAG) technology to generate context-aware answers using only the information found in the uploaded PDFs. ChatPDF provides a simple, user-friendly web interface for document ingestion and question answering, powered entirely by local resources—no internet-based APIs are required after setup.

---

🔧 Technologies Used

• Streamlit – For building the graphical web interface
• LangChain – Manages interaction between LLMs, embeddings, and retrieval
• ChromaDB – Local vector database for storing and retrieving document chunks
• Ollama – Hosts Large Language Models (LLMs) like Mistral locally

---

1. System Requirements

• Operating System: Windows 10/11, macOS, or Linux
• Python: Version 3.10 or higher
• Memory (RAM): Minimum 4 GB
• Storage: At least 5 GB available disk space
• Internet: Only required for setup (to install packages and download the model)

💡 How to open Command Prompt / Terminal:

• Windows: Press Windows + R, type cmd, and press Enter
• macOS: Use Spotlight (Cmd + Space) and search for "Terminal"
• Linux: Open the Terminal from your application menu or press Ctrl + Alt + T

---

2. Project Folder Structure

ChatPDF_Project/
├── main.py           # Application backend
├── app.py            # Streamlit frontend
├── requirements.txt  # Python dependencies
└── vectorstores/     # Auto-generated when app runs

---

3. Installation Procedure

3.1. Install Python 3.10+

1. Download from: python.org/downloads
2. During installation, check the box for "Add Python to PATH"
3. Verify installation:

   python --version

   It should return Python 3.10 or higher.

---

3.2. Set Up a Virtual Environment (Recommended)

In your project folder:

# Create a virtual environment
python -m venv venv

# Activate the environment
# On Windows:
venv\Scripts\activate
# On macOS/Linux:
source venv/bin/activate

💡 You should see (venv) appear before your prompt, meaning the virtual environment is active.

---

3.3. Install Project Dependencies

Create a file named requirements.txt with the following content:

streamlit
streamlit-chat
langchain
langchain-community
langchain-ollama
langchain-chroma
chromadb
pypdf
jq

Then install:

pip install -r requirements.txt

---

3.4. Install Ollama and Run the Mistral Model

1. Download and install from ollama.com/download
2. Then run:

   ollama run deepseek-r1:1.5b
   ollama run nomic-embed-text

   This will download (~1.4 GB) and start the Mistral model locally.

---

3.5. Add the Application Code

Ensure your folder has these two files with the following content:

• main.py – [Full backend logic]
• app.py – [Streamlit frontend interface]

📁 Make sure these are in the same directory as your requirements.txt.

---

4. Running the Application

From the project directory (and with the virtual environment activated):

streamlit run app.py

Your browser will open automatically or show a local link (e.g., http://localhost:8501).

---

5. Using the Application

1. Click Upload PDFs to ingest your documents
2. Type a question and click Send
3. View document-aware answers in the chat
4. Use 🗑️ Delete All Data to reset and clear vector memory

---

🔄 What's New in v2.1.0

• ✅ General JSON Support: Ingest both defectList[] (e.g., MRT damage logs) and category[].elements[] (e.g., masonry catalog).
• ✅ Prompt Improvements: Short, clear, paragraph-only answers. Avoids repetition, bullets, and numbering.
• ✅ Flexible Retrieval: Loosened similarity threshold for better JSON matching.
• ✅ Cleaner UI Flow: Auto-detection of PDF vs. JSON during upload; graceful error handling.
• ✅ Automatic Truncation: Ensures concise 3-sentence answers even if the model overruns.

---

📄 License

This project is licensed under the MIT License.