https://github.com/xeven777/eve.ai
Eve.AI is a system that learns communication patterns from WhatsApp chat history and creates a personalized AI chat experience that mimics a specific person's unique communication style, language patterns, and personality.
https://github.com/xeven777/eve.ai
ai gemini gemini-ai gemini-api langchain whatsapp
Last synced: 7 months ago
JSON representation
Eve.AI is a system that learns communication patterns from WhatsApp chat history and creates a personalized AI chat experience that mimics a specific person's unique communication style, language patterns, and personality.
- Host: GitHub
- URL: https://github.com/xeven777/eve.ai
- Owner: Xeven777
- Created: 2025-01-21T17:02:03.000Z (9 months ago)
- Default Branch: main
- Last Pushed: 2025-03-03T20:16:49.000Z (7 months ago)
- Last Synced: 2025-03-03T21:25:15.260Z (7 months ago)
- Topics: ai, gemini, gemini-ai, gemini-api, langchain, whatsapp
- Language: TypeScript
- Homepage: https://eveai.netlify.app/
- Size: 1.03 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# 🤖 Eve.AI - Personalized Chat AI
Eve.AI is a system that learns communication patterns from WhatsApp chat history and creates a personalized AI chat experience that mimics a specific person's unique communication style, language patterns, and personality.
Youtube video : [EVE-AI](https://www.youtube.com/watch?v=xQorrCjTvJI)
The system consists of three main components:
1. **🧹 Data Processing Pipeline**: Cleanses and structures raw WhatsApp chat exports
2. **🧠 Vector Database**: Stores conversations for semantic retrieval
3. **⚙️ API Server**: Handles chat requests and generates personalized responses
## 🔍 How It Works
### 1. Data Processing
The system takes raw WhatsApp chat exports and processes them to create clean, structured data:
- Removes system messages, media notifications, and other noise
- Creates a structured JSON with participants and conversations
- Filters out very short messages and keeps meaningful interactions
```bash
# Process a WhatsApp chat export
npx tsx refine.ts
```
### 2. Vector Database Seeding
The processed conversations are embedded and stored in a Pinecone vector database:
- Text is split into manageable chunks
- Google's AI creates vector embeddings for each chunk
- These vectors are stored in Pinecone for semantic search
```bash
# Seed the vector database with processed chat data
node seed.js
```
### 3. API Server
The API retrieves relevant messages and generates personalized responses:
- When a query is received, it performs semantic search to find relevant past conversations
- These conversations provide context about how the person would typically respond
- A carefully crafted prompt instructs the language model to mimic the person's style
- The AI generates a response that matches the person's unique communication patterns
## 🚀 Setup Instructions
1. **Environment Setup**
Create a `.env` file with the following variables:
```
PINECONE_API_KEY="your-pinecone-api-key"
PINECONE_HOST="your-pinecone-host"
PINECONE_ENVIRONMENT="your-pinecone-environment"
PINECONE_INDEX="your-pinecone-index-name"
GOOGLE_API_KEY="your-google-api-key"
GROQ_API_KEY="your-groq-api-key" # Optional alternative model
NAME="Person's name"
FEATURES="Description of person's communication style and personality traits"
```
2. **Install Dependencies**
```bash
pnpm install
```
3. **Export and Process Chat Data**
Export your WhatsApp chat (see [WhatsApp Export Guide](./whatsapp-export-guide.md) for detailed instructions).
Place your WhatsApp chat export in the api directory as `demochat.txt` and run:
```bash
npx tsx refine.ts
```
4. **Seed Vector Database**
Ensure your Pinecone database is set up, then run:
```bash
node seed.js
```
5. **Start the API Server**
```bash
node index.js
```
## 📡 API Usage
### Chat Endpoint
```http
POST /chat
Content-Type: application/json
{
"question": "Ei, ki korchis?"
}
```
Response:
```json
{
"answer": "Ei kaj shesh kore ektu cooking niye busy chilam😆. Tui bol, tui ki korchis?"
}
```
## 🛠️ Technology Stack
### Backend
- **🔄 Runtime**: Node.js
- **🌐 Server Framework**: Express.js
- **🧠 AI/ML**:
- **LangChain**: Framework for building LLM applications
- **Google Gemini 1.5 Pro**: Primary language model
- **Groq/Llama 3.3 70B**: Alternative language model
- **Google text-embedding-004**: Embedding model for semantic search
- **💾 Database**:
- **Pinecone**: Vector database for efficient semantic retrieval
- **📝 Languages**:
- **JavaScript/TypeScript**: Core programming languages
### Development Tools
- **📦 Package Management**: npm/pnpm
- **⚙️ Environment Management**: dotenv
- **🔄 TypeScript Compiler**: tsx for direct execution of TypeScript files
## ✨ Customization
The personality and response style are controlled through environment variables:
- `NAME`: The name of the person being mimicked
- `FEATURES`: Description of the person's communication style and personality traits
> **Note**: This project is for educational purposes only and should not be used to impersonate or deceive others. Always respect privacy and ethical boundaries when working with personal data.
> This doesn't ensure that the AI will always generate responses that are 100% accurate to the person's style. The AI is trained on a limited dataset and may not always reflect the person's true communication patterns. And the accuracy of the responses may vary based on the quality of the training data and the language model used. Better paid models like Gpt4 or Claude 3.7 will provide more accurate results.