Add support for open source models

[bodhish]

The current version uses OPEN AI models to transcribe the audio and OAI LLM to transform the text, it would be great if we can add options to use open source LLM's and transcription models.

[selamanse]

@bodhish I'd like to try this if it is still of interest. can you assign me please?

[selamanse]

also some questions:

• did you have any specific llms in mind that you already now could be suitable?
• how do I test your service, do you have integration tests?
• I've roughly looked into he code and can see that the transctriptions controller is taking care of the transcription to form by using fill_form function of the openai api. can you point out other AI interaction relevant sections in the code?

[bodhish]

Hi @selamanse! I was travelling, sorry for the late replay.

1. For LLM recommendations:

   • Transcription: WhisperX/Faster-Whisper/Self-Hosted Whisper
   • Form Processing: Llama/Mistral

2. The project doesn't have a proper automated testing workflow, I will add a few for the services in the coming week.

3. The main AI interactions happen in:

• OpenaiHelper module (currently handles both transcription and completion)
• transcriptions_controller.rb (workflow orchestration)

I'd suggest extracting our AI interactions into a provider-based pattern that would allow easy swapping of models based on environment configuration that will let users pick a model for transcription and another for completion.