Hi @radames,
We are using a multi-agent architecture to source and generate the newsletter. Some key agents are:
- Sourcing agent: we fine-tune an OSS model to help understand the instruction about where to crawl the news and what information we’d like to add to the newsletter
- Ranking agent: pick up the most important and most relevant documents
- Generation agent: generation final text and image content of the newsletter.
There are multiple models involved in different agents, are for some agents like ranking and sourcing agents, we used huggingface pre-trained models.
Hope this can be helpful!
Thanks!