How to Stream LLM Responses in Django and React Using Server-Sent Events

A development team building AI features in Django initially used standard POST requests, causing users to wait up to 8 seconds for a full response before anything displayed on screen. Streaming was adopted as the solution, using Server-Sent Events (SSE) over WebSockets since LLM communication is one-directional from server to client. On the backend, Django's StreamingHttpResponse paired with a generator function yields tokens incrementally as they arrive from the LLM API. A critical production detail is setting the X-Accel-Buffering: no header to prevent Nginx from buffering the stream and negating the effect. On the frontend, the browser's native EventSource API consumes the SSE stream without requiring additional dependencies.