A multi-platform chatbot that provides intelligent customer and tech support using OpenAI, Dialogflow, and a Retrieval-Augmented Generation (RAG) system.
Table of Contents
Features
- Multi-platform Support: Messenger, Telegram, LINE, Instagram, or custom web page.
- Retrieval-Augmented Generation (RAG): Intelligent, document-based responses using semantic search.
- Dynamic Intent Handling: Dialogflow for intent matching and tagging documents for improved efficiency.
- Context-aware Responses: Combines RAG and OpenAI models for enhanced conversational AI.
- Document Processing: Upload, chunk, and store documents with embeddings for semantic search.
- Multiple AI Model Support: OpenAI, Mistral, META (Llama), Together AI, with configurable AI Provider Switching.
Demo
How It Works
- Intent Handling:
- Dialogflow matches user inputs with known intents and tags uploaded documents.
- Only documents matching specific tags are searched to improve efficiency.
- RAG Process:
- Uploaded documents are chunked with overlapping sections.
- Embeddings are generated and stored for semantic search.
- Relevant chunks are retrieved using a weighted combination of cosine similarity and fuzzy matching scores.
- Retrieved context is added to prompts for response generation using GPT models.
Persistent Conversation Context:
- Persistent Conversation Context:
- Redis stores user conversation history in key-value pairs, allowing personalized, context-aware responses.
- History is fetched and included in prompts for OpenAI and Dialogflow, ensuring continuity across interactions.
- Cross-platform Integration:
- APIs for Messenger, LINE, Telegram, and Instagram.
- Custom web frontend built with React.
- Language Model Selection & Switching:
- The chatbot supports multiple Language Models (OpenAI GPT-4, Mistral-large, META Llama-3, Together AI).
- Only one model is active at a time, while Dialogflow can be enabled/disabled independently for intent matching.
- Users can switch between models using commands (e.g.,
/openai, /mistral, /meta).
Tech Stack
- Frontend: React
- Backend: Go (Gin framework)
- Database: PostgreSQL
- In-memory Store: Redis (for conversation history and context storage)
- APIs: OpenAI, Dialogflow, Telegram, LINE, META, Together AI, Hugging Face
- Cloud Deployment: Google Cloud (backend), GitHub Pages (frontend)
- Tools: PDF processing libraries for text extraction
Installation
-
Prerequisites:
-
Install Python Packages:
- The backend requires several Python packages for processing PDFs. To install them, run:
pip install pdfplumber pytesseract pdf2image PyPDF2
- These packages handle PDF text extraction and Optical Character Recognition (OCR) for images within PDFs.
-
Set Up Environment Variables:
- Create a
.env file in the configs/ directory to store variables such as API keys and database configurations (refer to sample.env).
-
Run the Backend Server:
- Navigate to the project directory and start the server:
go run main.go
Frontend Setup (Optional)
Potential Use Cases
- Streamlining FAQs and troubleshooting for customer support.
- Automating knowledge retrieval for internal teams.
- Enhancing collaborative workflows with intelligent document processing.
- Acting as a meeting assistant for document organization and context provision.
- Extending to incorporate custom-trained or pretrained language models.
License
This project is licensed under the MIT License - see the LICENSE file for details.
If you have any questions or suggestions, feel free to reach out to [email protected].
评论