Backend Software Engineer — Node.js · TypeScript · Python

I build systems where correctness under concurrency and failure matters more than surface features — financial ledgers that can't silently lose data, multi-provider AI pipelines that degrade instead of breaking, and background job systems designed to survive crashes and races rather than just handle the happy path.

Open to Global Remote Roles & Contractor Contracts, remote or relocation.

Node.js TypeScript Express PostgreSQL Redis BullMQ React Supabase

A multi-tenant platform for tracking shared family money and responsibilities — recurring contribution pools, disputes, tasks — built the way a fintech backend would be, not a typical CRUD app. 199 API endpoints, ~25–30 core tables, 18 documented Architecture Decision Records.

• Financial integrity by construction — confirmed ledger entries are never edited in place; corrections are new linked entries, so history is always reconstructable. Idempotency keys plus a 10-minute duplicate-submission heuristic guard against both network retries and genuine double-submits.
• Atomicity via Postgres RPCs — dispute resolution, invite acceptance, contributor-target updates, and event→recurring-pool conversion all run as single-transaction stored procedures, closing check-then-write race windows that sequential application code can't safely close.
• 9-queue BullMQ system with per-queue concurrency tuning and a Postgres-backed notification outbox — failed push/email deliveries are marked, not dropped, and a recovery scan re-enqueues anything stuck.
• Layered authorization — JWT → workspace membership → role, with a 30-second Redis membership cache (actively invalidated on writes) and enumeration-resistant 404s instead of 403s on unauthorized workspace access.
• Real test coverage — unit, integration, and rate-limit suites, with CI running against actual PostgreSQL and Redis rather than mocks.

🔗 Live demo

Node.js TypeScript Express PostgreSQL Redis BullMQ

A coaching platform where real outreach outcomes and AI-simulated practice conversations are scored on the same rubric and feed one coaching loop.

• 4-provider LLM fallback chain (Cerebras → Groq → Mistral → OpenRouter), with failures classified by structured HTTP status/error body — a bad key, a down provider, and a dead model each trigger a different recovery action instead of one generic retry.
• AI cost gating — a dedicated module decides whether an AI call is worth making before spending it (research reuse, low-stakes skip, quota checks), with every decision logged to an audit table.
• One bundled AI call instead of four for practice-session replies — reply text, buyer internal monologue, outcome classification, and coaching feedback all come back from a single request.
• Redis-coordinated cross-instance state — key cooldowns and model-discovery caching are shared across horizontally-scaled instances, with a documented kill switch back to in-memory behavior.

Python Flask PostgreSQL Redis WebSockets

A peer academic collaboration platform (group chat, live study sessions, AI tutoring) built to run correctly under multiple worker processes, not just a single instance.

• Redis-coordinated multi-instance state — WebSocket presence, rate limiting, and scheduled-job locking all moved off process-local memory so gunicorn -w N is actually safe.
• Fail-closed distributed locking — the one deliberate exception to an otherwise fail-open Redis usage pattern, because a missed scheduler tick is cheaper than a duplicate cron run.
• 6-provider AI fallback with Redis-backed health state and the same status-code-driven failure classification approach as FounderSales.
• Latency isolation — AI dispatch runs on a bounded 8-worker thread pool so a slow model call can't block the WebSocket event loop handling real-time messages.
• Refresh-token rotation with reuse detection — a 10-second grace window distinguishes a legitimate multi-tab race from an actual stolen-token replay.

🔗 Live demo

Freelance Web Developer — local salon booking platform (Jun–Aug 2025) Designed and shipped a Node.js/Express + PostgreSQL booking API end-to-end — requirements through deployment (Vercel/Railway, custom domain + SSL) — and maintained it in production for 3 months post-launch.

oluwaseyiogunsola90@gmail.com · LinkedIn · Portfolio