Data architecture
Vector DBs, structured APIs, and fine-tuning each pull additional hardware. Fine-tuning gets its own GPU pool — it's not part of inference capacity.
Customer Support RAG
bge-large=1024 · OpenAI ada=1536 · e5-mistral=4096
Fine-tuning