Live Commerce Marketplace

Context

A regional e-commerce client needed a social commerce platform where sellers could list products, stream live shows, run real-time auctions, and receive wallet-based settlements — all in one app. No existing codebase existed. The product needed to reach production on both iOS and Android with a matching web client.

Constraints

• The entire product — mobile app, backend, and web version — had to be built and shipped by one engineer.
• Financial flows (top-up, checkout, settlement, payout, reversal) required correctness guarantees that a typical CRUD approach could not provide.
• Live streaming and auction state had to stay consistent across hundreds of concurrent mobile clients without a dedicated real-time backend.
• The app needed Arabic and English support with multi-environment release flavors from day one.

Approach

I designed a feature-first Flutter architecture with Riverpod state management, typed Supabase repositories, and AutoRoute navigation guards. Financial flows were modeled as an explicit ledger with processing → settlement → payout lifecycle, implemented in TypeScript Edge Functions. Live streaming used Agora RTC for video with Supabase Realtime channels for auction state synchronization across viewers.

Key decisions

• Chose Supabase as the backend platform to own auth, Postgres schema, RLS policies, storage, and edge functions from one place — avoiding a separate backend service that one engineer could not maintain.
• Modeled wallet transactions as an append-only ledger rather than balance mutations, so reversals and audits stayed clean without touching settled records.
• Kept auction bid state in Supabase Realtime rather than Agora data channels, so auction logic remained server-authoritative and auditable.

Tradeoffs and risks

• Accepted Supabase Realtime latency (~100–200ms) for auction bids instead of a lower-latency custom WebSocket server, trading absolute speed for operational simplicity on a solo build.
• Deferred automated test coverage (9 test files in ~149k LoC) to hit the launch deadline, with the plan to add coverage to the settlement and reversal paths post-launch.
• Used Supabase Edge Functions for all financial logic despite cold-start constraints, to avoid managing a separate Node.js service.

Outcomes

• Shipped a full production Flutter app (~149k lines of hand-written code, 1,015 files under lib/) on iOS and Android within 6 months as a solo engineer.
• Shipped a parallel Next.js web version of the same platform.
• Financial ledger handled buyer payments, seller settlement cron, referral commissions, and transaction reversals without data integrity issues in production.
• Live streaming and real-time auction flows ran stably across concurrent viewers using Agora RTC + Supabase Realtime.

Artifacts

Technology