Ask HN: Help me understand RAG vs. fine tuning for building a coding partner

The author, a Ruby on Rails enthusiast with limited time due to personal and professional commitments, is interested in creating a Rails coding assistant. This AI-powered tool would generate code based on user requirements, allowing the user to review, refine, and test the code before accepting it. The author is unsure about the best architecture for this tool, whether to use RAG (Retrieval-Augmented Generation) or fine-tune a model with all the content. They have found a project, Augmentoolkit, which seems to align with their idea but are still uncertain.

Additionally, the author is considering investing in a maxed-out Mac Studio for training and self-hosting, with a budget of $10k. They are seeking advice on whether this would be a good investment for their project. The author also acknowledges the time commitment required to build such a tool and asks readers to disregard copyright concerns for the sake of discussion.

Key takeaways

The author has experience with Ruby on Rails and is interested in creating a Rails coding assistant that can generate code based on user input.
The author is considering using RAG (Retrieval-Augmented Generation) or fine-tuning a model with content to create this tool.
The author found a project called Augmentoolkit that seems to align with their vision but is unsure if it's the right choice due to frequent mentions of RAGs.
The author is contemplating investing in a maxed out Mac Studio for training and self-hosting, with a budget of $10k.

Ask HN: Help me understand RAG vs. fine tuning for building a coding partner

Key takeaways

Discussion (0)