> For the complete documentation index, see [llms.txt](https://voxrad.gitbook.io/voxrad/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://voxrad.gitbook.io/voxrad/fundamentals/getting-set-up/understanding-workflow.md).

# Understanding Workflow

VOXRAD uses two ways to transcribe audio to report.

1. Use a combination of using a transcription model to first transcribe audio and then format and restructure the transcript using instruction template.
2. Use a multimodal model to directly input the audio and instruction template to provide output.

{% hint style="info" %}
**Important:** If using the first method, you will have to provide two keys, i.e., for "Transcription Model" and "Text Model" in the settings. For using the second method only a single "Multimodal Model" key is required.
{% endhint %}

## Supported LLMs

There are 3 types of LLMs supported by the application.

<table><thead><tr><th width="180">Model</th><th>Capabilities</th></tr></thead><tbody><tr><td>Transcription Model</td><td>Transcribes audio to text. Models like <code>whisper</code>. Most API services have an upper limit of 25 MB.</td></tr><tr><td>Text Model</td><td>Utilizes the transcript and instruction template to generate response. Models like <code>gpt-4</code> and <code>Llama 3</code>.</td></tr><tr><td>Multimodal Model</td><td>Can directly use user's recorded audio and instruction template to generate output. Models like <code>gemini-1.5-flash</code></td></tr></tbody></table>

{% hint style="warning" %}
Only `gemini-1.5-pro` and `gemini-1.5-flash` multimodal models are supported experimentally. `GPT-4o` will be supported as the API becomes available. As these are remotely hosted LLMs should not be used for any sensitive data.
{% endhint %}