🧠 RagFlow

RAGFlow is a clean, modular, production-style conversational AI application built with
Retrieval-Augmented Generation (RAG).

It enables document-grounded conversations, multi-chat sessions, streaming responses, persona selection, and multilingual output — with a cleanly separated backend and frontend.

🖼️ Application Preview

✨ Key Features

🏗 Clean Architecture

• Fully separated backend and frontend
• Backend can run independently (CLI today, API-ready for the future)
• Clear service boundaries for maintainability and scaling

📄 Retrieval-Augmented Generation (RAG)

• Upload documents: PDF, TXT, DOCX, CSV
• Automatic text extraction, chunking, and embedding
• FAISS-based vector search for fast, relevant retrieval
• Per-chat document isolation (no context leakage)

💬 Chat Experience

• Multi-chat sessions (create, switch, rename)
• Each chat maintains its own history and context
• Token-by-token streaming responses for real-time UX

🎭 Personalization & Language

• Built-in personas:
  • Friendly Assistant
  • Formal Expert
  • Tech Support
• Multilingual responses:
  • English
  • Urdu
  • Arabic

🗂️ Project Structure

ragflow-chat/
├── backend/                       # Core backend (frontend-agnostic)
│   ├── __init__.py                # Package marker
│   ├── config.py                  # Central config (chunk sizes, model names, etc.)
│   ├── model.py                   # LLM + embeddings initialization (Gemini)
│   ├── graph.py                   # LangGraph workflow / orchestration
│   ├── chat_service.py            # High-level streaming chat service (UI/CLI call this)
│   ├── rag.py                     # Retrieval service (query → relevant context)
│   └── document_rag.py            # Document ingestion: extract → chunk → embed → FAISS
│
├── frontend/                      # Streamlit frontend (UI only)
│   ├── streamlit_app.py           # UI entrypoint (wires sidebar + chat UI)
│   ├── sidebar.py                 # Sidebar UI: chats list, RAG upload, settings, actions
│   ├── chat_ui.py                 # Chat rendering + streaming response display
│   └── state.py                   # Session-state helpers (new chat, rename, safe indexes)
│
├── scripts/                       # Backend-only utilities (no Streamlit required)
│   ├── chat_cli.py                # CLI chat (persona + language; no document RAG)
│   └── rag_cli.py                 # CLI RAG: loads a local document path, then Q&A
│
├── assets/
│   └── ui.png                     # Screenshot used by README (optional but recommended)
│
├── .streamlit/
│   └── secrets.toml               # Local API keys (DO NOT COMMIT)
│
├── requirements.txt               # Python dependencies
├── .gitignore                     # Ignore venv, secrets, caches, etc.
└── README.md                      # Project documentation

🧰 Tech Stack

• Python 3.11+
• Streamlit (UI)
• LangChain + LangGraph (orchestration)
• Google Gemini (LLM + embeddings)
• FAISS (vector search)
• PyPDF2 / python-docx / pandas (document parsing)

⚙️ Setup

1) Create a virtual environment

python -m venv venv

Activate:

• Windows (PowerShell)

  .\venv\Scripts\Activate.ps1

• macOS / Linux

  source venv/bin/activate

2) Install dependencies

pip install -r requirements.txt

3) Add secrets

Create:

.streamlit/secrets.toml

GOOGLE_API_KEY = "YOUR_GOOGLE_API_KEY"

This file is ignored by Git and should stay local.

▶️ Run the Streamlit UI

streamlit run frontend/streamlit_app.py

Open:

• http://localhost:8501

💬 Run Chat CLI (backend-only)

Runs the chatbot in your terminal (no Streamlit).

python scripts/chat_cli.py

🧪 Run RAG CLI (backend-only)

This runs RAG entirely in your terminal.

python scripts/rag_cli.py

How “document upload” works in the CLI

In rag_cli.py, you provide a local file path when prompted (that’s the “upload” step for CLI).
Supported formats: pdf, txt, docx, csv

The CLI will then:

1. Read the file from disk
2. Extract text
3. Chunk text (overlapping chunks)
4. Create embeddings
5. Build an in-memory FAISS index
6. Let you ask questions grounded in the document

Notes

• Vector store is in-memory (not persisted)
• Each run creates a fresh index

🚀 Future Improvements

• Persist vector stores (disk/DB) per user/chat
• Add FastAPI backend for web/mobile clients
• Authentication + user accounts
• Dockerization / deployment templates
• Better document parsing (OCR for scanned PDFs)

👤 Author

Built by Sajid Ali
Focused on clean architecture, RAG systems, and maintainable AI applications.