Design a Social Marketplace for Products

Problem statement

Design a social marketplace where individuals list products for sale, other users follow sellers and browse a personalized feed, like and comment on listings, negotiate over direct messages, and finalize the purchase either through an in-app payment or an external checkout link. Think Etsy's creator economy, Depop's follow-first social model, and Facebook Marketplace's local discovery, compressed into one product aimed at local and niche communities rather than a global everything-store.

The brief requires four functional pillars: posting items with images and prices; liking, commenting, and following; direct messaging for negotiation; and payment or an external link to close the sale. It also requires four quality attributes: scalability for user-generated listings, search and filtering by category and location, trust and safety with scam reporting, and integration with shipping or in-person meetup.

Why this problem is distinctive

A plain e-commerce site optimizes a catalog that is effectively infinite and seller-agnostic. A social marketplace has three properties that change the architecture:

1. Listings are finite, unique physical items. In consumer-to-consumer resale there is exactly one unit of each listing. Two buyers can click "buy" on the same vintage jacket at the same second, and the system must produce exactly one winner without losing either buyer's trust. Inventory correctness here is closer to ticketing than to Amazon-style retail.

1. Discovery is social before it is transactional. Users follow sellers, like items, and comment. The feed is the storefront. That means the read-heavy social machinery — fan-out, ranking, counters — sits on the critical path of revenue, exactly as it does at Instagram or Twitter, except that every impression can convert into a one-of-one sale.

1. Trust is the product. Strangers trade with strangers. Scams, counterfeit goods, off-platform payment solicitation, and no-show meetups destroy the network from the inside. Moderation, reputation, and dispute handling are not features; they are the substrate the marketplace runs on.

Public operating baseline versus design assumptions

Public evidence shows the category operates at serious scale. Meta reported more than one billion people using Marketplace monthly as of 2021. Vinted reported passing 100 million registered members across its European C2C fashion marketplace in 2024. Etsy's 2023 reporting described roughly 90 million active buyers and about 7.5 million active sellers. Mercari reported over 20 million monthly active users in Japan at peak. These are company-reported figures and context only — not requirements for our fictional system.

For capacity planning this answer explicitly assumes a mature niche network: 30 million registered users, 6 million DAU, 20 million active listings, 1.2 million new listings per day, and a four-times event peak for feed reads and search. Unless tied to a citation, every number is a stated design assumption, target, or budget.

The four architectural planes

1. Social plane: follows, feed generation, likes, comments, counters, and ranking. Read-heavy, latency-sensitive, eventually consistent.

1. Commerce plane: listings, images, categories, inventory state, offers, checkout, and payouts. Correctness-critical, strongly consistent per aggregate.

1. Messaging plane: negotiation conversations, delivery guarantees, presence, and spam control inside DMs. Per-conversation ordering, at-least-once with idempotent application.

1. Trust plane: listing moderation, image screening, scam detection, reports, appeals, enforcement, and reputation. Asynchronous, evidence-driven, human-in-the-loop.

A strong interview answer keeps these planes separate. The social plane may degrade to stale counters without hurting checkout. The trust plane may slow risky listings without stopping trusted sellers. And the commerce plane never lets a stale social cache decide who owns the only unit of an item.