Why it matters
Embedding and retrieval choices directly determine how much context gets stuffed into a prompt, so understanding the mechanics here is a prerequisite for controlling RAG-driven token costs.
The tokenmaxxing angle
Cosine similarity retrieval is the mechanism that decides how much (and how relevant) context an agent pulls into its prompt, making this talk a direct look at one of the biggest levers on RAG token spend.
From the organizers
Speaker Dev Shah live-codes a RAG pipeline in the session, showing embeddings stored in a vector database and retrieved in real time via cosine similarity.