Backpressure
Backpressure is a solo-built educational load simulator for LLM-serving and system-design work. Users assemble an architecture on a canvas, run deterministic simulated load through it, and see bottlenecks, queueing behavior, latency, errors, and the breaking point before production users do.
At a Glance
- ML and platform engineers sizing inference infrastructure
- Backend and systems engineers
- Developers and teams reviewing architecture
- AI coding-agent and developer-tool users
- +1 more
AI Tools by Backpressure
(1)
Backpressure
LLM Infrastructure Load Simulator
Discussions
No discussions yet
Be the first to start a discussion about Backpressure
Latest News
Show HN: backpressure.systems adds MCP and CLI
Leonardo Cardoso shared how he built Backpressure’s interactive editor for dragging components, connecting architectures, and pushing load through them.
Cardoso presented Backpressure as the system-design tool in his emerging AI-agent toolkit and previewed MCP support.
Published “Honest Math for LLM Serving,” a technical explanation of Backpressure’s KV-cache, queueing, deterministic simulation, and failure-point models.
Products & Services
A canvas-based load simulator for system-design and LLM-serving architectures. Users place and connect modeled components such as traffic sources, gateways, queues, caches, databases, GPUs, models, and KV caches, then run load to receive a survived/collapsed verdict, bottleneck, metrics, and breaking point.
An assistant grounded in the live board that explains why a design failed or which component became the bottleneck in plain language.
A local stdio MCP server that lets AI coding agents list components, create/read/update/delete boards, run boards, and simulate designs without saving them.
An npm-installed thin client for the same API, with authentication, component catalog access, architecture simulation, and CI behavior that can fail a build when a design collapses or a pinned bottleneck moves.
Market Position
Backpressure positions itself between static capacity calculators and conventional architecture diagrams: it is an interactive, deterministic simulator that emphasizes queues, overload, bottlenecks, tail latency, and the exact load at which a design breaks. Its LLM-serving angle specifically challenges optimistic GPU calculators by modeling KV-cache concurrency and decode limits rather than only completions per second.
Leadership
Founders
Leonardo Gomes Cardoso
Senior software engineer based in Berlin with more than a decade working across backend, web, and mobile systems. He currently works on the Cards team at Trade Republic; previously he spent three years at Blinkist on a product reaching millions of readers. His own sites describe Backpressure as a side project built off-hours.
Founding Story
Cardoso started Backpressure after seeing that free GPU/LLM capacity calculators reported optimistic completion-rate numbers rather than the real concurrency and KV-cache limits that cause serving rigs to fail. The initial vision was a canvas-driven simulator that would make system behavior and overload failures tangible: build a design, push load through it, watch it hold or collapse, and learn where to fix it.
Business Model
Revenue Model
Freemium subscription model: a Free tier supports limited simulations and selected exports/features, while Pro provides unlimited simulations and advanced capabilities. Backpressure also advertises a one-time lifetime Founder deal for the first 100 purchasers.
Pricing Tiers
2 simulations; 100 AI credits per month; JSON export, Skye AI, live metrics, remix, MCP server, and CLI. Load sweeps, comparisons, PDF/spec/Mermaid exports, multiplayer, live updates, and other advanced capabilities are excluded.
Unlimited simulations; 1,000 AI credits per month; load sweeps, design comparison, all listed exports, Skye conversation memory, board invites, multiplayer, live updates, MCP server, and CLI.
One-time lifetime deal for the first 100 people: Pro forever, future Pro features, Founder badge, founders wall, direct access, early access, and roadmap input.
Target Markets
- ML and platform engineers sizing inference infrastructure
- Backend and systems engineers
- Developers and teams reviewing architecture
- AI coding-agent and developer-tool users
- Students and candidates preparing for system-design interviews
- Sizing and stress-testing LLM inference rigs before buying hardware
- Learning system design and queueing behavior through interactive failure
- Evaluating backend request paths involving gateways, caches, queues, replicas, and databases
- Investigating thundering herds, retry storms, pool starvation, load shedding, and bottlenecks
- Architecture regression tests in CI via the CLI
- System-design interview preparation