Backpressure is a web-first, interactive load simulator for system design and LLM serving. You build a system on a canvas, apply real concurrent load, and watch where it breaks - queue depth, time-to-first-token, throughput, and the moment it saturates and collapses. A deterministic queueing engine backs every run, so the same design and the same load always produce the same result, and each run is a shareable, reproducible artifact.
The launch focus is LLM serving: assemble a serving setup (model, quant, GPU or hardware, batching), simulate concurrent users, and watch time-to-first-token and tokens per second degrade until the rig recovers or falls over. The same engine and visual language extend to general system design - load balancers, queues, caches, and datastores under pressure.
The interactive canvas needs JavaScript. Explore Backpressure without it here: