Any suggestions on managing document chunks and vector storage in RAG setups? Do you recommend combining them or keeping them separately?
imho, an integrated set up works fine for small scale, but when data grows, seperating doc chunks and vectors cuts update issues. i noticed this approach lowers conflicts and makes scaling smoother.
i reckon its simpler to keep them seperate so you can update each independently. merging them sometimes mixes things up, esp when doc sizes vary. this way, each system gets optimized without conflict
In my experience, maintaining separate systems for document chunks and vector storage can simplify updates and troubleshooting. Separating these components allows independent scaling, which is particularly useful when document sizes vary significantly or when the vector database requires frequent refinements. However, I have also experimented with integrated approaches that reduce complexity in query routing and data coherence. Ultimately, the optimal architecture depends on specific requirements such as update frequency and system workload. Thoughtful modular design can enhance both performance and maintainability in retrieval-augmented generation systems.
hey, i wonder if a hybrid approach might work well – merging can simplify some queries, but sometimes update issues occur. curious to hear if anyone faced similar hiccups? what made you choose one way over the other?