README.md

metadata

license: afl-3.0
title: RAG
sdk: docker
emoji: 📚
colorFrom: green
colorTo: yellow

TechFlow Customer Support Agent 🤖

A robust, RAG-based AI agent designed to assist users by answering questions from local PDF documentation and seamlessly creating GitHub support tickets when answers are unavailable.

🚀 Live Demo

Try it on Hugging Face Spaces

🌟 Features

• RAG (Retrieval-Augmented Generation): Answers user queries using a local knowledge base created from PDF documents.
• Smart Search & Synthesis:
  • Typo Tolerance: Handles misspellings (e.g., "solutiun") and infers user intent.
  • Relevance Filtering: Automatically discards search results with a similarity score below 0.7 to prevent hallucinated answers from irrelevant text.
  • Comprehensive Summarization: Synthesizes answers from multiple document chunks.
• Strict Citations: Every answer includes mandatory source citations with filenames and page numbers (e.g., **Source:** manual.pdf (Page 12)).
• Automated Ticket System:
  • Condition-Based: Offers to create a ticket only when the agent explicitly "could not find the answer".
  • GitHub Integration: Directly creates issues in a specified GitHub repository.
  • Project Organization: Automatically applies labels and prefixes titles based on the configured PROJECT_FOLDER.
• Auto-Ingestion: Automatically checks and updates the vector database from the data/ directory when the application starts.

🛠️ Prerequisites

• Python 3.9+
• OpenAI API Key
• GitHub Personal Access Token (with repo scope)

🚀 Installation & Setup

1. Clone the Repository

   git clone <your-repo-url>
   cd <project-directory>
   

2. Install Dependencies

   pip install -r requirements.txt
   

3. Configure Environment Create a .env file in the root directory (copy from .env.example):

   cp .env.example .env
   

   Update .env with your credentials:

   OPENAI_API_KEY=sk-...
   GITHUB_TOKEN=ghp_...
   REPO_NAME=your_username/your_repo_name
   PROJECT_FOLDER=if_ur_project_has_folder_name_put_it_here  # Used for ticket labels/titles
   

4. Add Knowledge Base Place your PDF documents into the data/ folder.

   • The system will automatically ingest them when you run the app.

💻 Usage

Start the Streamlit application:

streamlit run app.py

How it works:

1. Ask a Question: Type your query in the chat.
2. Get an Answer: The agent searches the PDFs.
   • If a high-confidence answer is found (Score > 0.7), it answers with citations.
   • If relevant info is NOT found, it replies "I could not find the answer".
3. Raise a Ticket: If the answer was not found, a form appears allowing you to create a GitHub issue immediately.

• app.py: Main Streamlit application entry point.
• agent.py: Core logic for the AI agent, RAG search, and TicketSystem.
• ingest.py: Script handles loading PDFs, chunking, and FAISS vector storage.
• data/: Directory for source PDF files.
• vector_db/: Directory where the FAISS index is saved locally.