sathishaiuse/Salesforce-Voice-Insights

Salesforce Voice Insights — Open Source Bundle

This bundle contains a Streamlit app that:

• —Records or accepts audio.
• —Transcribes audio locally using faster-whisper.
• —Generates insights using a local LLM (llama-cpp-python) if available, otherwise a simple fallback summary.
• —Uploads the audio and a Note (transcript + insights) to a Salesforce record.

What's included

• —app.py — Streamlit application.
• —requirements.txt — Python dependencies.
• —Dockerfile — Containerize the app.
• —download_model.sh — Helper script with instructions to fetch models.
• —huggingface/ — placeholder folder for downloaded model files (not included due to size).
• —README.md — This file.

How to run (local)

1. 1.Install dependencies:

   python -m venv venv
   source venv/bin/activate
   pip install -r requirements.txt

1. 1.Put Whisper model files under models/ (see download_model.sh).

1. 1.(Optional) Put a llama/ggml model under models/ and set the path in the app sidebar.

1. 1.Run the app:

   streamlit run app.py

How to run with Docker

docker build -t sf-voice-insights .
docker run -p 8501:8501 --rm -it sf-voice-insights

Notes on Models and Licensing

• —Model weights are not included in this bundle. Download them yourself and ensure you comply with the model's license.
• —faster-whisper can use Hugging Face model snapshots or the whisper.cpp repository. For best performance, use GPU where available.

Extending and Customizing

• —Replace the fallback insights with any other local LLM backend.
• —Add OAuth flow for Salesforce instead of pasting session ID.
• —Improve query logic for related objects and fields per your Salesforce schema.