Dudil

Dudil is an emotion-aware chatbot that analyzes user input using a pre-trained emotion classification model and responds empathetically through Google’s Gemini AI. The application is designed to create contextual, emotionally intelligent conversations through a clean Streamlit-based interface.

---

Features

• Emotion Detection: Classifies text into emotions such as joy, love, sadness, anger, fear, and surprise.
• Empathetic AI Responses: Dynamically adapts responses based on emotional state.
• Conversation History: Persists chats across sessions with support for browsing and management.
• Confidence & Intensity Scoring: Provides insight into how strongly emotions are detected.
• Streamlit UI: Modern web interface with expanders, metrics, and chat components.

---

Tech Stack

• Frontend: Streamlit
• Emotion Model: Hugging Face Transformers (DistilBERT/DistilRoBERTa variants)
• Conversational Model: Google Gemini API
• Storage: JSON-based chat history
• Language: Python 3.7+

---

Prerequisites

• Python 3.7 or higher
• Internet connection (for model/API)
• Google Gemini API Key

---

Installation

1. Clone the repository

   git clone https://github.com/your-username/dudil.git
   cd dudil

2. Install dependencies

   pip install -r requirements.txt

3. Configure environment variables Create a .env file:

   GEMINI_API_KEY=your_gemini_api_key
   GEMINI_MODEL=gemini-pro
   DISTILBERT_MODEL=j-hartmann/emotion-english-distilroberta-base

4. Obtain Gemini API Key

   • Visit Google AI Studio
   • Generate a key and add it to .env

---

Usage

1. Run the application

   streamlit run app.py

2. Interact

   • Go to http://localhost:8501
   • Enter your message
   • View real-time emotion analysis and contextual replies

---

How It Works

1. User input is analyzed using an emotion classification model.
2. The detected emotion and intensity are used to construct a contextual prompt.
3. Gemini AI generates a response tailored to the emotional context.
4. Results are displayed with emotion metrics and stored in history.

---

Supported Emotions

Emotion	Intensity	Description
Joy	5/5	Happiness and positive tone
Love	5/5	Affection and care
Surprise	4/5	Sudden or unexpected reaction
Anger	2/5	Frustration or irritation
Fear	2/5	Anxiety, nervousness
Sadness	1/5	Melancholy, disappointment

---

Environment Variables

Key	Description
GEMINI_API_KEY	Required for accessing the Gemini API
GEMINI_MODEL	Gemini model name (e.g., gemini-pro)
DISTILBERT_MODEL	Hugging Face emotion classification model

---

Model Configuration

• Temperature: 0.7
• Top-p: 0.8
• Top-k: 40
• Max Tokens: 1024

---

Chat Features

• Create and manage new conversations
• View and delete previous chats
• Clear all chat history
• Auto-save on session exit

---

Troubleshooting

Model Loading

• Ensure stable internet for downloading models
• Check disk space (~500MB required)

API Connection

• Verify that your API key is valid
• Ensure you have quota and correct permissions

Chat History

• Ensure chat_history.json is writable
• Corrupted files are backed up automatically

---

Contributing

1. Fork this repository
2. Create your branch: git checkout -b feature/my-feature
3. Commit your changes: git commit -m 'Add new feature'
4. Push to your branch: git push origin feature/my-feature
5. Open a pull request

---

License

This project is licensed under the MIT License.

---

Credits

• Hugging Face
• Google AI
• Streamlit
• j-hartmann – Emotion Model