This project is a WhatsApp AI Assistant configurable for your needs. It allows you to create a personalized AI chatbot for your business. The Service is set to use a language model on backend of WhatsApp chatbot and be configurable from Admin Panel. It also has a Telegram bot for managers, a CRM system, an option to edit all prompts, vector database for searching items, prompt construction based on results from vector database, automatic deletion of outdated data to the database, Prompt construction based on results from vector database, Reminder mechanism to bot, and more.
The project consists of 3 services:
The bot is used to communicate with clients, consult them on basic questions about your business, and report on predefined set of cases to the backend
The panel is used to configure the bot's behavior. Specifically:
- Basic instructions in form of natural language
- Items representing elements on which users ask questions
- Scenarios to report to backend
- Other configurations
The bot can be added to group or channel, and it will report on activations of triggers described above. Managers who serve triggered clients are stored in leader board that can later be viewed to check who did what
The video with demo of basic functionality.
WhatsApp bot is running on account +7 (964)-860-22-14 Admin Panel can be accessed here Telegram bot can be accessed here
- Clone repository
git clone https://gitlab.pg.innopolis.university/n.zagainov/ai-sales.git
- Go to working directory
- Install Requirements
npm ci
- Additional installation
npm install -g pm2
This is needed to run the project
- Go to Firebase console
- Create new project or choose existing one
- In your project configurations, add firebase realtime database
- Create
.envfile and copy credentials from firebase console:
API_KEY="YOUR_API_KEY"
AUTH_DOMAIN="YOUR_AUTH_DOMAIN"
PROJECT_ID="YOUR_PROJECT_ID"
STORAGE_BUCKET="YOUR_STORAGE_BUCKET"
MESSAGING_SENDER_ID="YOUR_MESSAGING_SENDER_ID"
APP_ID="YOUR_APP_ID"
MEASUREMENT_ID="YOUR_MEASUREMENT_ID"
DATABASE_URL="YOUR_DATABASE_URL"- Go to Green API console, create an instance for your bot
- Copy ID and API key to
.env:
ID_INSTANCE="YOUR_ID_INSTANCE"
API_TOKEN_INSTANCE="YOUR_API_TOKEN"- Go to Google AI Studio, generate API key and choose model (e. g.
gemini-1.0-pro), paste configurations to.env:
GEMINI_TOKEN="YOUR_API_TOKEN"
GEMINI_MODEL="gemini-1.0-pro" # replace if needed
EMBEDDING_MODEL="embedding-001" # replace if needed- Add proxy to
.envto workaround Gemini limitations:
PROXY_URL="YOUR_PROXY_URL"- Add telegram bot token to
.env:
TELEGRAM_TOKEN="YOUR_TELEGRAM_TOKEN"- Go to Pinecone, create new index named
ai-saleswith 768 dimensions and cosine similarity criterion, add your API token to.env
PINECONE_TOKEN="YOUR_PINECONE_TOKEN"- Set port for Admin Panel in
.env
PORT=8000 # change if neededTo run the project, type:
npm run start
This will run all the services: WhatsApp Bot, Admin Panel, Telegram Bot. To stop:
pm2 stop all
- Edit `src/admin-panel/add-admin.js and run
npm run add-admin
to add new authorization to admin panel
- Go to port your host (if running locally,
http://localhost:<YOUR_PORT>/) - Configure System prompts for definition of behavior, intent detection, and reminder
- Create new intents (e. g: "I want to buy", "I want to speak to the manager")
- Optionally, add items on which your bot will assist users
- Go to the chat with your bot
- Test your configurations
- To add trigger activation tracking, add telegram bot to the group of choice and type
/set_group - Bot will stream activated triggers to this group and manage services made by admins