# System design

> Architecture answers with diagrams.

## What it does

The System Design tab takes a design X or scale Y question and returns a structured answer with capacity estimates, functional + non-functional requirements, scalability tiers, an architecture diagram, trade-offs, edge cases, and deep-dives. The Scale Calculator lets you adjust DAU / QPS inline; scalability tiers are clickable drill-downs that re-stream a focused answer for any specific component (Cassandra, Kafka, Redis, etc.).

## Asking design questions

Phrase the question with a recognized verb — design X, build X, how would you scale X. The phrasing nudges Cariara into design mode rather than the general inference path. Include constraints up front: read-heavy or write-heavy, latency target, region count. The capacity estimate uses whatever you give it; vague inputs produce vague numbers (which you can then refine with the Scale Calculator). The Design tab accepts the same multi-input modes as Coding: paste, fetch from URL, drag-drop screenshot, or voice via the bottom-bar mic.

## Sections you’ll see

The answer streams back as a sequence of cards. Each one renders independently and fills in token-by-token, so you can start reading early sections before the deep-dives finish. One-paragraph framing of the system being designed. A pitch / walkthrough — how you’d talk through this on a whiteboard, in order. Functional + Non-Functional Requirements Two side-by-side cards. Non-functional usually covers latency, availability, consistency, scale targets. Scale Estimates + Scale Calculator Back-of-envelope DAU, QPS, storage, bandwidth — with an interactive calculator that lets you tweak the inputs and recompute. Scalability Tiers Each tier is a clickable card — click to re-stream a focused answer for that specific tier (e.g. "tell me more about Cassandra here"). Hard decisions and the rationale behind them — the kind of content interviewers grill you on. Edge Cases Failure modes, races, retry storms, partial outages. Two or three of the hardest sub-problems answered in detail. Where most of the points are scored in a real interview.

## Scale Calculator

Inside the Scale Estimates card you’ll find the Scale Calculator — sliders / inputs for daily active users, requests per second, payload size, and retention. Adjust any of them and the downstream numbers (storage, bandwidth, server count) recompute live. Use this to react to interviewer questions like what if it was 10x bigger? in real time without re-asking Sona.

## Scalability tier drill-downs

The Scalability Tiers card lists the components that scale your system — database tier, cache tier, queue tier, etc. Click any tier and Cariara re-streams a focused answer for just that component (Cassandra vs DynamoDB vs CockroachDB, with the trade-offs for your scenario). Useful when an interviewer drills into one piece without asking you to redesign the whole system.

## The architecture diagram

A rendered architecture diagram appears below the prose, typically a few seconds after first tokens arrive. It’s purely visual — every component shown is also listed in the prose, so the answer is complete with or without the image. Diagrams are cached at the problem level. The first reader on a fresh question pays the generation cost; every subsequent visitor sees the same image instantly. Click Regenerate on the design header to bypass the cache and ask for a fresh answer. A small fraction of design questions don’t yield a clean diagram — usually because the generated code tripped the safety check or the renderer hit its time budget. The prose answer is unaffected. Rephrase more concretely (specific components, specific cloud) and resubmit.

## Detail level

Toggle between Basic and Full in the design header. Basic shows the headline components and reasoning; Full adds supporting services (caches, queues, observability) and longer deep-dives. Pick Basic for a 30-second sanity check, Full for a 5-minute deep dive. The cloud-provider icon set used in the diagram is configured globally in Settings (not per-question) — Cariara picks AWS by default; switch to GCP or Azure in Settings if you prefer those icons.

## Sona follow-up sidebar

The right rail on the Design tab is a Sona sidebar — a focused Q&A pane that knows the active design context. Ask follow-ups like why Cassandra not DynamoDB here? or how would you handle a cross-region failover? without leaving the tab. Different from the live-interview Sona stream — this one is for deliberate exploration.

## Tips for live sessions

Submit the moment you’ve heard the question. The architecture diagram needs a few seconds to render — early submission means it’s ready when you start drawing on your own whiteboard. Talk through the Functional Requirements card first. Interviewers love candidates who clarify scope before diving into components. Use the Scale Calculator live during the call — if the interviewer says what if DAU went up 10x? , change the input and the storage / bandwidth / server-count numbers update under your eyes. The Tradeoffs and Deep-dives sections are where most of the points are scored. Use them as the script for tell me more about how X works follow-ups. If the interviewer drills into one tier, click that tier’s card to get a focused answer instead of re-asking the whole question. Live session setup, audio capture, and Cmd+B safety are documented in Live Session .

Source: https://docs.cariara.com/live-design