Design scalable backend systems with expert trade-offs

A persona skill that acts as a senior system-design engineer across five operating modes selected from context: DESIGN produces a full architecture blueprint, REVIEW audits an existing system for flaws, SCALE produces a phased growth roadmap, INTERVIEW simulates a system-design interview with scoring, and EXPLAIN answers conceptual questions, asking one clarifying question first if the mode is ambiguous. DESIGN mode runs a fixed six-step sequence: ask clarifying questions (use case shape, user counts, latency and availability targets, geographic spread, budget, existing stack) before designing anything; compute scale estimates explicitly, requests per second at average and peak, storage per day and year, inbound and outbound bandwidth, read-write ratio, cache-hit target, rather than describing scale in words; produce a twelve-layer architecture blueprint covering client/CDN, DNS and load balancing, API gateway, application-layer service decomposition, caching, database with an explicit SQL-vs-NoSQL decision and sharding strategy, message queues, object storage, search, observability, security, and CI/CD; render a Mermaid diagram customized to that specific design, never a generic placeholder; produce a technology-stack table with a reason per row; and state every major decision as an explicit decision, why, trade-off, and alternative block. REVIEW mode audits a described system using ten fixed tags, single point of failure, bottleneck, scale limit, security gap, data-loss risk, latency issue, cost inefficiency, observability gap, tight coupling, antipattern, and reports findings grouped into critical, high-priority, and medium-priority sections, plus what the system already does well. SCALE mode produces a four-phase roadmap, MVP/startup, growth, scale, hyper-scale, each phase naming the trigger metric to advance, what to build versus buy, and an estimated monthly infrastructure cost range. INTERVIEW mode runs a seven-step flow, problem statement, clarifying questions, scale estimation, high-level design, a deep dive on 2-3 components, a bottleneck-at-10x-scale discussion, then scoring, and produces a six-category scorecard out of 30 with a hire or no-hire verdict. Fourteen named design patterns and a technology decision matrix are applied and justified automatically when relevant, with deeper reference files for patterns, technology comparisons, scale benchmarks, a security checklist, and cost-estimation formulas.

Use it for architecting, reviewing, or scaling backend systems - designing a new system's architecture, auditing an existing one for weaknesses, planning how a system should evolve as usage grows, practicing a system-design interview, or getting a conceptual explanation of a distributed-systems concept. Its own output standards require every response to justify each component choice, compute real numbers instead of describing scale vaguely, state trade-offs for every decision, proactively flag risks even in DESIGN mode, produce a customized diagram every time, give a phased roadmap unless a single phase was explicitly requested, take an opinionated recommendation rather than listing options without a pick, name antipatterns when a request implies one, and reason about what happens when each component fails. Its stated limitation: as AI-generated design guidance, it can be incorrect, so a design should be verified before it goes to production.

Input is a system-design request, an existing architecture description, a target user count, or a request to practice an interview. Output depends on mode: a DESIGN blueprint (clarifying answers, scale-estimation math, the twelve-layer architecture, a Mermaid diagram, a tech-stack table, and trade-off statements); a REVIEW audit report grouped by severity with tagged findings; a SCALE roadmap across four growth phases; or an INTERVIEW transcript ending in a scored, six-category verdict.

References external technology categories throughout its recommendations, CDNs, load balancers, API gateways, Redis/Memcached, relational and NoSQL databases, message queues, object storage, search engines, and observability and IaC tooling, without integrating with any of them directly - it is a design and review persona, not an execution tool - and points to its own bundled reference files for deeper pattern, technology, benchmark, security, and cost detail.

Engineers and teams who want a structured, numbers-driven system-design partner - for designing a new architecture, auditing an existing one, planning a scaling roadmap, or practicing a system-design interview - rather than a vague, unjustified list of technology options.