AmritaGPT

AmritaGPT is a chatbot designed to answer all Amrita Viswa Vidyapeetham related questions, covering topics such as clubs, placements, entrance exams, and more. The system facilitates text-to-text conversation as well as speech-to-text and text-to-speech functionalities. 🤖📚🎙️

Prototype

Implementation

• AmritaGPT collects data from various sources including the Amrita website, Quora, and other relevant platforms. This data is utilized by a RAG-based Large Language Model (LLM) to generate responses. The project relies on LLAMA 3/Gemini model, FAISS, and Langchain for its functionality. The API gateway is powered by FastAPI, with temporary use of ngrok. 🌐🛠️

• For speech-to-text conversion, Whisper is employed, which sends the input to the LLM. gTTS is then used to convert the generated text into speech. 🔊🗣️

• The front-end is developed using React.js, while Flask is used for backend operations. The web UI and system integration are currently under development and will be completed shortly. 💻⚙️

Chatbot API runninng

This Section explains how to set up, run, and interact with the chatbot API.

---

Prerequisites

1. Python: Install Python 3.9 or higher.
2. Dependencies: Ensure the required Python packages are installed.
3. Environment File: Create a .env file and add your gemini and huggingface key.
4. Text Data: Ensure a text file named general.txt exists in the root directory containing the knowledge base.
5. Models:
   • HuggingFace sentence-transformers/all-MiniLM-L6-v2
   • Google Generative AI Embeddings (embedding-001)
   • Meta Llama Llama-3-8B-Instruct

---

Installation

1. Clone the repository:

   git clone <repository_url>
   cd <repository_directory>

2. Install dependencies:

   pip install -r requirements.txt

3. Configure the .env file:

   GOOGLE_API_KEY=<your_google_api_key>
   HF_API_TOKEN = <your token>

---

Running the API

1. Start the server:

   python api.py

2. The API will be available at:

   http://127.0.0.1:8000
   

---

Endpoints

POST /get-response/

• Description: Get chatbot response.

• Request Body:

  {
      "session_id": "<optional_session_id>",
      "input_text": "<user_question>",
      "use_google": false
  }

  • session_id (optional): Reuse a session ID for conversation continuity.
  • input_text: The user query.
  • use_google: Use Google Generative AI (true) or HuggingFace (false).

• Response:

  {
      "session_id": "<session_id>",
      "response": "<bot_response>",
      "history": [
          {"user": "<input_text>"},
          {"bot": "<response_text>"}
      ]
  }

---

Features

• Chat History: Maintains context from the last two exchanges.
• Embedding Models: Supports both HuggingFace and Google Generative AI embeddings.
• Custom Prompts: Tailored for educational use cases.

---

Customization

1. Change Models:

   • Update the model_name for HuggingFace embeddings in:

     huggingface_embeddings = HuggingFaceEmbeddings(model_name="<new_model_name>")

2. Modify Prompt:

   • Adjust the prompt template in get_conversational_chain() to fit your use case.

3. Add New Endpoints:

   • Use FastAPI’s routing capabilities to add more endpoints as needed.

---

Troubleshooting

1. Model Loading Errors:

   • Ensure all required models are correctly placed in the models directory.

2. Environment Variables Not Found:

   • Check that .env is correctly configured and loaded.

3. API Not Starting:

   • Ensure all dependencies are installed and use Python 3.9+.

---

Contact

For issues, please contact Team IETE.

Acknowledgements

This project is developed by IETE Amrita SF under the initiative of IETE Amrita SF 2023-24 team, Amrita Vishwa Vidyapeetham, Coimbatore.

© 2024 IETE Amrita. All rights reserved.