Daniel Nashed's Blog

If you’re interested in RAG, you should watch the OpenRAG Submit video

Niklas Heidloff published a short blog post that pointed me to it https://heidloff.net/article/openrag/
That post was the trigger to look at this excellent OpenRAG Submit video.


OpenRAG Submit
https://www.youtube.com/watch?v=y3Tr1i_ynvI


Even if you never plan to use OpenRAG, this video is absolutely worth to watch.
What makes it stand out is that it explains RAG by breaking it down into individual components, instead of treating it as a black box. It focuses on architecture, responsibilities, and trade-offs rather than on a single framework or product.

This is not just an OpenRAG video — it’s one of the clearest explanations of modern RAG architectures I’ve seen so far.
The speakers are clearly experts in their fields, but just as importantly, they are excellent communicators. The presentation is structured, easy to follow, and genuinely enjoyable to watch.
And yes — I  love the Run MCP T-shirt.

My Takeaways from the video
 

• Many tools used in AI, RAG, and related components are open source
  That makes perfect sense. Openness, transparency, and extensibility are essential in this space.
  This also helps to accelerate development in the AI space
  
  
• RAG does not automatically mean “vector database”
  Vector databases are useful, but they are not mandatory. The video does a great job of explaining alternative approaches and when they make sense.
  
  
• MCP servers are an important part of the stack
  They play a key role in how components communicate and integrate.
  
  
• Data ingestion and tokenization really matter
  How data is ingested — and how it is tokenized — has a major impact on results. Tokenization strategies change over time and should be flexible and pluggable.
  
  
• Everything should be pluggable
  All components in a RAG architecture should be replaceable. One of the most important emerging standards enabling this is MCP.
  

• Comments [0]