Backend & Platform Engineer

Today: TypeScript throughout. Expo / React Native (SDK 54, expo-router) mobile; Next.js 16 web; Node/Express with Prisma API; Socket.IO realtime; Stripe Connect destination charges; Checkr background checks; Firebase and Expo push. Target: AWS — ECS or App Runner, RDS PostgreSQL with PostGIS, S3; Google Gemini 2.0 Flash for job classification; Google Places and Distance Matrix. None of the AWS layer exists yet and the database is still SQLite. The role You will take a working development backend and turn it into something that can take real money from real people in New York. The API is built and covers most of the domain — 15 route groups, 17 data models, Stripe Connect, Checkr, Socket.IO. What it does not have is a production environment. There is no cloud deployment, no container setup, no CI/CD, and the database is still SQLite. That migration and that infrastructure are the job. What you will do • Migrate SQLite to PostgreSQL. The Prisma schema is written against SQLite with a note that Postgres is the target. You will do the migration properly — real enums, nullable uniques, indices, and a repeatable seed and migration path. • Stand up production on AWS. Nothing exists today — no Dockerfile, no CI/CD, no infrastructure-as-code. You will containerise the API onto ECS or App Runner, provision RDS and S3, and build the deploy pipeline, logging, metrics and alerting from scratch. • Own the money end to end. Stripe Connect is wired for connected accounts, onboarding links and destination charges with an application fee. The lifecycle is missing: authorise on booking with manual capture, capture on confirmed completion, then a seven-day payout lag with a ledger of funds held versus owed. Our balance sheet carries helper payables as a liability and nothing currently produces that number. Plus webhooks, idempotency, refunds, disputes and daily reconciliation against Stripe. • Own the geo layer, and its bill. Matching shortlists with haversine in application code — deliberately, because it is free and instant. You will move that filter to PostGIS ST_DWithin on a GIST index, then call Google Distance Matrix for travel time on the final shortlist ONLY. Maps bills per element at traffic-aware rates; running it across a whole borough would cost roughly twenty-eight times our entire hosting budget. Caching repeated origin-destination pairs is part of this job. • Verification pipeline. Checkr is integrated for helper background checks. You will own the lifecycle — webhooks, re-checks, document expiry, and the compliance side of who is allowed to take which category of job. • Security and data protection. Card data never touches our servers, so our PCI scope is SAQ-A and should stay that way. You will own secrets management, access control, encryption of personal data, and backup and restore that has actually been rehearsed. • Build the async layer. Everything runs inline in the request today. Gemini calls, Checkr polling, payout batches and notification fan-out all need a queue — SQS or BullMQ. Socket.IO also needs a Redis adapter and sticky sessions before it can run on more than one ECS task. • Put tests in. There is no test suite. You will introduce one, plus CI that blocks merges on it. What we need • 5+ years backend engineering in Node.js and TypeScript. Go or Python elsewhere is fine. • You have run a real PostgreSQL migration and can talk about what went wrong. • You have built production cloud infrastructure and CI/CD from nothing, not just maintained someone else's. • Hands-on Stripe Connect, or an equivalent marketplace payments and payouts system. • Comfortable with webhooks, idempotency and the ways distributed payment flows fail. Nice to have • PostGIS and spatial indexing for dispatch or delivery. • AWS specifically — ECS or App Runner, RDS, S3, SQS. • Prisma specifically. • Marketplace escrow, split payments or delayed payouts.