What would you say are the key differences between your own custom RAG pipeline implementation vs using OpenAI GPTs or HuggingFace Assistants?

My 2 cents for the former:
1) cost —> by using open source llm
2) accuracy —> document preparation and the processing optimized for the task at hand
3) speed?

Systems thinker and product guy. Founder of   App in the Air. Now bulding Wingman prediction market (wingman.wtf) and sharing growth hacks at t.me/web3_growth

overall, if you are building RAG application, I'd love to exchange experience, please DM