e-johnstonn / salescopilot

Intelligent sales assistant built using Deep Lake, Whisper, LangChain, and GPT 3.5/4

SalesCopilot 🚀💸

SalesCopilot is an AI-powered sales assistant that provides real-time transcription, live chat, automatic unprompted advice, integration with a custom knowledge base, and the ability to save and load past conversations.

Built with Deep Lake, LangChain, and modified versions of ecoute and Speech Recognition. All open-source, check them out!

Features

• Real-Time Transcription: Transcribes your conversations with in real-time, maintaining a record in the 'Transcript' tab for review and analysis.
• Live Chat: Ask questions, get advice, and more with a chat bot that reads and understands the live transcript.
• Unprompted Advice: Potential objections or questions the customer has are detected, and advice on how to respond is offered within seconds.
• Knowledge Base Integration: Uses Deep Lake as a vector database to store and retrieve information, allowing your chosen sales guidelines to be queried, with the most relevant being used to give advice.
• Save and Load Transcripts: Save transcripts, then load them up later and have it summarized, ask for a performance evaluation, and more.

Demo (sound on)

demo_vid.mp4

Setup 🔧

• As of now only Windows is supported. A separate branch supporting macOS is coming soon!
• Ensure you have FFmpeg installed. If you don't, here's a guide.

1. Clone the repository and navigate to the project directory

git clone https://github.com/e-johnstonn/SalesCopilot.git
cd SalesCopilot       

2. Install required packages: pip install -r requirements.txt
3. Set your OpenAI API key in keys.env
4. By default, audio will be transcribed using the Whisper API. If you have an NVIDIA GPU and want to transcribe locally, set USE_API to False in AudioTranscriber.py, and install torch with CUDA

Running SalesCopilot

To start the program, python main.py in the project directory.

Enter a name for the person you're speaking to, click Start, and the app will load. In the "Sales Assistant" tab, you can chat with the GPT 3.5 powered Sales Assistant. This is also where advice regarding detected objections will appear.

When the conversation is finished, you can click "Save and Quit" in the "Transcript" tab to save a copy of the transcript. If you restart the app, you can now load this transcript. Ask SalesCopilot to summarize it, evaluate your performance, or any other questions related to the transcript.

Using your own knowledge base

By default, the app uses this as a knowledge base, located in the data folder. To use your own knowledge base:

1. Put your knowledge base in the data folder in the form of a text file
2. Update the path in chat_utils.py to the path of your knowledge base
3. I recommend adjusting the split_data method in deep_lake_utils.py to split your document effectively - if it's unstructured use something LangChain's RecursiveCharacterSplitter

How the knowledge base works:

License

MIT License

🤝 Contributing

Contributions are welcome!

salescopilot's Issues

Support for models other than OpenAI?

Hi there,

Just wondering if you plan to add support for LLMs other than just OpenAI's GPT 3.5/4?

This project looks really interesting, thanks for sharing! :)

Using Nova instead of Whisper

Are we able to use Nova API instead of Whisper API? It's supposed to be much faster, accurate and cheaper.

https://blog.deepgram.com/nova-speech-to-text-whisper-api/?utm_source=twitter&utm_campaign=0423%20nova%20whisper%20announcement&utm_medium=organic%20social&utm_content=announcement%20tweet

Just wondering what would need to be done to support Nova API being used here.

Transcription Failed Error

I get transcription failed error. I have installed the requirements.txt file and entered my API key both with and without the <>'s to test.