@bashifuirkashi: RAG vs CAG explained 👇 Most people learning AI engineering know RAG. But CAG, Cache Augmented Generation, is another approach you should understand. With RAG: → Documents are chunked → Chunks are turned into embeddings → Embeddings are stored in a vector database → A user query is embedded → Similar chunks are retrieved → The LLM generates an answer using that retrieved context With CAG: → You preload the documents directly into the model’s context → The model creates a KV cache → User queries are added to that context → The LLM answers using the information already loaded Sounds simpler, but CAG has tradeoffs. You still have to think about: ⚙️ Context window limits ⚙️ Scalability ⚙️ Cost ⚙️ Latency ⚙️ Accuracy ⚙️ Data freshness Knowing how to build a RAG demo is one thing. Knowing when to use RAG, CAG, or another retrieval architecture is what starts moving you toward production-level AI engineering. Comment “CAG” and I’ll DM you the link to my AI engineering community with: ✅ A full AI engineering learning roadmap ✅ Daily calls with working AI/ML engineers ✅ Recruiter and career guidance ✅ Projects, interview prep, and everything you need to work toward landing a $150K+ AI engineering role

Vanessa Eyenga :

Instant follow !! 🙌🏾

AFK :

CAG is what claude and chatgpt do by default how is this helpful

ShearQuery :

Which One is cheaper

christoph_hugo :

So Cag is basically context stuffing.

War Badger :

How can this plausibly make economic sense, paging intermediate states for faster response loading sounds good until you realize that the ratio from actual kv-cache to document size is astronomical. Maybe if this is like a constant thing for a small set of caches you’ve reserved but it’s not even plausibly an alternative to RAG embeddings because embedding models are a fraction of the size and several orders faster then the models they do retrievals for.

Alren :

Rag kinda useless. But I have it on my resume

Legend of Lore :

Wow thanks for the breakdown! I didn’t even know CAG was a thing until now.

ima81natio9 :

And it costs a ton of tokens every time you load the same documents into the context

Cowries :

CAG

Crisco_Alpha :

This is good

Tensor Space :

I am serving a CAG system. I think it is cool that you are convering this. I go with CAG whenever I can here is my SaaS: tensorspace.net

Safiya Stephenson-Andrews :

CAG

steve_nomen :

CAG

Rockon :

cag

Marc Platvoet :

cag 🥺

Lucia.G.tiktok :

cag

Jim :

cag

Naqo :

CAG

Cisco562 :

CAG

Kenny Mcspenny :

I’m only still at the ingestion stage

asp_uk :

Cag

ofir_sharfi :

CAG

NubianShehzadi5 :

CAG

Filipe Tarouca :