Design Decentralized Exchange

Problem Statement: Permissionless Spot DEX

A decentralized exchange lets anyone swap ERC-20 tokens without a custodian or an order book — liquidity sits in on-chain pools and price is computed by a formula, not matched against resting orders. The canonical design is a Uniswap/Curve-class automated market maker (AMM): traders swap against a pool, the pool's reserves move along a constant-product curve x · y = k (or a stable-swap curve for pegged assets like USDC/USDT), and liquidity providers (LPs) earn the swap fee for supplying the reserves. The contracts are the product; your backend is the read-and-route layer around them.

The scale and economics to anchor: a major DEX holds billions in TVL (Uniswap has run $3–8B+), charges a 0.3% fee on v2-style swaps, and settles every swap atomically on-chain. The defining boundary, exactly as in the other DeFi questions: settlement truth lives on-chain (the pool contract is authoritative), while the indexer/API projections that power quotes and charts may lag a block. You never let an off-chain quote authorize a trade the on-chain swap would revert.

The core invariant that makes the whole thing safe: k never decreases within a swap except for the fee that accrues to LPs — the contract checks reserves after the token transfer and reverts if the product would shrink. That single post-condition is what stops a malformed or malicious swap from draining a pool.

The failure stories that define a DEX (and what an interviewer is really probing) are AMM-specific: a sandwich attack front-running and back-running a large trade with tight slippage, fee-on-transfer or rebasing tokens breaking naive balance accounting, indexer lag showing a stale price-impact preview, and a reorg double-applying a Swap event in your off-chain projection. These, not a generic microservices diagram, are the substance of the role.