Automated Customer Support Chatbot

A next-generation hybrid customer support chatbot combining the power of Rasa 3, LangChain, Llama.cpp, and FastAPI, with a sleek Streamlit chat UI. Deliver both structured and generative answersleveraging local LLMs and Retrieval Augmented Generation (RAG)for exceptional customer experiences.

---

Features

• Hybrid Intelligence: Rasa's robust dialogue + LLM-powered generative answers
• Local LLaMA Backend: Fast, private, and cost-effective (supports unsloth.Q8_0.gguf and more)
• RAG Support: LangChain + ChromaDB for document-grounded answers
• FastAPI Microservice: Scalable LLM API for custom actions
• Modern Streamlit UI: Chat-style, responsive, and cloud-ready
• Flexible Deployment: Run locally, on your server, or deploy the UI to Streamlit Cloud

---

Architecture

mermaid flowchart TD A[" User"] --> B[" Streamlit Chat UI"] B --> C[" Rasa 3 Server"] C -- "Custom Action (action_ask_gpt)" --> D[" FastAPI LLM API"] D --> E[" Llama.cpp LLM"] D --> F[" LangChain RAG (ChromaDB, Docs)"]

---

Project Structure

text Automated-Customer-Support-Chatbot/ actions/ # Custom Rasa actions (calls LLM API) data/ # Rasa NLU, stories, rules models/ # Trained Rasa models, LLaMA GGUF files results/ # Output, logs, etc. app.py # Streamlit frontend llm_api.py # FastAPI LLM+RAG backend start_chatbot.py # Script to launch all services locally requirements.txt # Python dependencies README.md # This file ... # Other configs and scripts

---

How It Works

1. User chats via the Streamlit web UI
2. Streamlit sends messages to the Rasa 3 server (REST API)
3. Rasa handles intent/entity recognition and dialogue
4. For open-ended/knowledge queries, Rasa triggers action_ask_gpt:
   • Calls the FastAPI LLM API
   • API uses LangChain to retrieve context (RAG) and generate a response with Llama.cpp
5. The answer flows back to the user via Rasa and Streamlit

---

Tech Stack

| Backend | LLM & RAG | Frontend | Deployment | |-----------------|-------------------|---------------|--------------------| | Rasa 3 | Llama.cpp | Streamlit | Local/Cloud | | FastAPI | LangChain | | Streamlit Cloud | | Python 3.83.10 | ChromaDB | | Docker (optional) |

---

Quickstart

1. Clone the Repository

sh git clone https://github.com/your-username/Automated-Customer-Support-Chatbot.git cd Automated-Customer-Support-Chatbot

2. Install Dependencies

sh pip install -r requirements.txt

3. Set Up Environment Variables

Create a .env file: env OPENAI_API_KEY=your_openai_key GGUF_MODEL_PATH=path_to_your_llama_model.gguf

4. Train and Start Rasa

sh rasa train rasa run --enable-api

5. Start the LLM API

sh python llm_api.py

6. Start the Streamlit Frontend

sh streamlit run app.py

Or use start_chatbot.py to launch all services together (locally).

---

Deploying on Streamlit Cloud

• Only the Streamlit frontend is deployed on Streamlit Cloud.
• Rasa and the LLM API must run on a separate server (local, VM, or cloud).
• Set the Rasa server URL in your Streamlit Cloud environment variables.

---

Customization

• Add Documents: Place PDFs, Markdown, or text files in the docs/ folder for RAG.
• NLU & Stories: Edit data/nlu.yml, data/stories.yml, and data/rules.yml for intents and flows.
• Custom Actions: Extend logic in actions/.
• Model: Swap out LLaMA models by changing GGUF_MODEL_PATH.

---

Troubleshooting

• See TROUBLESHOOTING.md for common issues and solutions.
• Key tips:
  • Ensure Rasa and LLM API are running and accessible
  • Check environment variables and API keys
  • Review logs for errors

---

License

MIT License

---

Made with by Faisal Hakimi