SalesCopilot is an AI-powered sales assistant designed to enhance sales performance through real-time conversation transcription, live chat support, and proactive advice. It targets sales professionals seeking to improve their interaction quality and outcomes by leveraging advanced AI models and a custom knowledge base. The primary benefit is providing immediate, context-aware guidance during sales calls.

How It Works

The system utilizes Whisper for real-time audio transcription and LangChain for orchestrating interactions with GPT 3.5/4. It features a live chat interface that analyzes the ongoing transcript to offer unprompted advice, detect potential customer objections, and suggest responses. A key component is its integration with Deep Lake as a vector database, enabling the retrieval of relevant information from a custom knowledge base (e.g., sales guidelines) to inform the AI's advice.

Quick Start & Requirements

• Install FFmpeg.
• Clone the repository and run pip install -r requirements.txt.
• Set your OpenAI API key in keys.env.
• For local Whisper transcription, set USE_API = False in AudioTranscriber.py and ensure PyTorch with CUDA is installed.
• Run with python main.py.
• Note: Currently Windows-only; macOS support is planned.

Highlighted Details

• Real-time transcription and live chat with GPT 3.5/4.
• Unprompted advice generation based on detected objections.
• Custom knowledge base integration via Deep Lake vector database.
• Functionality to save, load, summarize, and evaluate past transcripts.

Maintenance & Community

The project is open-source with contributions welcome. No specific community channels or notable contributors are listed in the README.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive for commercial use and integration with closed-source applications.

Limitations & Caveats

The application is currently Windows-only, with macOS support pending. The effectiveness of the knowledge base integration may depend on proper document splitting, requiring potential adjustments to the split_data method.