Customerify

A simple factory for customer service agents. Open up our application, specify your data sources (we take online websites documents)

Then, select a phone number to deploy your agent with. Once done, customers can simply call that phone number and get answers to their questions from our fast, lifelike customer service agents.

Built with

• Express JS - for our Twilio / Websocket endpoints
• Flask - for our Agent response api
• Streamlit - for our frontend
• Twilio - Voice call integrations
• Deepgram - STT
• Groq - LLM provider
• LMNT - TTS
• OpenAI - embeddings
• PineCone - Vector DB

Running our application:

To run our application, you need to have ngrok installed. You will also need Twilio, LMNT, Deepgram, OpenAI, PineCone, and Groq accounts.

First, run

ngrok https 8080

to start the ngrok server (handles forwarding to your machine). Next, go to your Twilio console and set the webhook url as https://<NGROK_URL_WITHOUT_HTTPS>/start. Make sure to replace <NGROK_URL_WITHOUT_HTTPS> with the url that ngrok is forwarding to your machine.

Next, run the below script to get all the components up and running

# api keys
export DEEPGRAM_API_KEY=<your_deepgram_api_key>
export TWILIO_API_KEY=<your_twilio_api_key>
export TWILIO_ACCOUNT_SID=<your_twilio_accound_sid>
export GROQ_API_KEY=<your_groq_api_key>
export PINECONE_API_KEY=<your_pinecone_api_key>
export LMNT_API_KEY=<your_lmnt_api_key>
export OPENAI_API_KEY=<your_openai_api_key>

export FLASK_URL=http://localhost:5000
export NGROK_URL=<ngrok_url_without_https://>
export PHONE_NUMBER="123 456 7890"  # currently, we mandate that the Twilio user provide the phone number they set up
cd server
npm i
npm start &
cd ..
cd ./flask-backent
flask --app api.py run &
cd ..
streamlit run ui.py

Finally, open up the UI and configure your agent. Press deploy and call your agent to see the results!