Improve the transcription and summarization using LLMs #130
Description
The current transcription workflow has a summarization feature that uses Deepgram and a correction flag that requires manual work.
I propose to automate these 2 tasks using an LLM (Openai, Gemini or any other model) to correct the transcripts and generate a summary of the corrected transcript.
These 2 operations can also be server as an API endpoint compatible with the current backend.
This can be integrated in the current workflow without any major changes.
After this integration, I will try to improve the correction of the LLM using a dictionary/lookup table that it'll use as context during correction. This can be somewhat related to issue #122
After all this we can have a RAG with the transcripts and summary that can be plugged into the Braidpool's context manager.