AI-native Full-Stack Developer  ·  React · TypeScript · Node · MongoDB

Full-stack developer with 2+ years building and shipping production-grade, type-safe web applications — I architect, build, and operate a live UK e-commerce platform end to end, solo. I care about clean feature-driven architecture, API contracts that break TypeScript at compile time (not runtime), and payment systems resilient enough to survive a forced processor migration with zero downtime.

• Currently building: an AI support/knowledge SaaS (RAG over MongoDB Atlas Vector Search + Gemini) and a real-time live-auction platform (Socket.io + Redis)
• Focus: React 19 · TypeScript · RAG pipelines · real-time systems · payment architecture
• Open to: full-stack roles (local + remote) and freelance work

React 19 · TypeScript · Node · Express 5 · MongoDB · TanStack Query v5 · Tailwind v4 · Square · Vercel · Railway

• Architected a provider-agnostic payment layer that absorbed two processor migrations (Stripe → Mollie → Square) with zero schema changes, containing a forced shutdown to a 4-day driver swap — storefront never down.
• Built and operate the entire three-service platform solo — an Express / MongoDB API plus a React storefront and admin dashboard — with subdomain-scoped HTTP-only-cookie auth, guest checkout, and real production orders.
• Engineered idempotent payment processing with retrying MongoDB transactions, an in-transaction PAID re-check, and a send-then-mark email pattern to prevent duplicate charges and confirmation emails under webhook replay.
• Enforced a feature-driven architecture across all three codebases with frontend types mirroring API responses, so any contract change breaks at TypeScript compile time rather than in production.
• Achieved Lighthouse Accessibility 100, SEO 100, Best Practices 100 across storefront routes, with desktop Core Web Vitals in Google's "Good" range (LCP 1.94s, INP 88ms, CLS 0.04).

Live →  ·  Case Study → (case study — source private)

React 19 · TypeScript · Google Gemini · Express 5 · Mongoose · Zod · Docker · GitHub Actions

• Shipped a full-stack AI app end-to-end — React + Express / MongoDB API — that turns Bengali YouTube cooking videos into editable, structured recipes via Google Gemini, for a real non-technical user on a $0 inference budget.
• Designed a fault-tolerant asynchronous job pipeline (fire-and-forget + client polling) with wall-clock timeouts, status-gated race guards, and startup orphan-recovery, so a hung or rate-limited AI call can never crash the server.
• Anchored the system on a single Zod schema serving four consumers (AI response, Mongoose model, API envelope, and client type), so backend changes break the frontend at compile time, not runtime.
• Owned split-origin deployment: multi-stage Docker image → GitHub Actions → GHCR → self-hosted box behind a Cloudflare Tunnel, with auto-deploying CI/CD, an atomic daily-budget guard, and per-IP rate limiting.

Live →  ·  Code →  ·  Infra Docs →

React 19 · TypeScript · Express 5 · MongoDB · TanStack Query v5 · Socket.IO · Zod · Tailwind v4 · Vercel · Railway

• Engineered a fully DB-driven RBAC layer resolved per request behind a single authorization guard with zero hardcoded role names, so editing a role's permission set takes effect on the user's next request — no re-login, no redeploy.
• Made multi-line sales transactional and race-safe with a Mongoose session and a guarded atomic stock decrement per line item, so two sales racing for the last unit resolve all-or-nothing with server-computed totals — a failed line rolls the entire sale back, never a phantom decrement.
• Solved first-party cookie auth across two different-origin deployments (Vercel + Railway) with a same-origin /api/* proxy and a single-flight refresh interceptor that queues concurrent 401s and replays each once — then authenticated the WebSocket, which can't ride that proxy, with an in-memory access token: one auth model, two transports.
• Delivered real-time low-stock alerts over Socket.IO scoped to a permission-gated room only product:update holders join, with the broadcast isolated from the committing sale transaction so a socket failure can never delay or roll back a sale.

Live →  ·  Code →

React 19 · TypeScript · Express 5 · MongoDB · TanStack Query v5 · Zod

• Modeled co-authorship as an interactive force-directed collaboration graph over a computed aggregation — deduping mirror author-pairs by a sorted canonical key and weighting each edge by shared-publication count.
• Structured a feature-driven full-stack architecture mirroring frontend features 1:1 to backend modules with a shared typed API contract, so any response-shape change breaks TypeScript at compile time, not runtime.
• Wired debounced keyword search spanning faculty and publications in a single round-trip, with case-insensitive backend matching, Zod-validated params, and highlighted matches showing why each result surfaced.

Live →  ·  Code →

More shipped work, live demos, and case studies on my portfolio →

What I use

• Frontend: React 19 · TypeScript · Tailwind CSS v4 · TanStack Query v5 · React Hook Form · Axios · Vite · Motion
• Backend: Node.js · Express 5 · MongoDB · Mongoose · REST APIs · JWT (HTTP-only cookies) · Zod
• Payments / Media / AI: Square · Stripe · Mollie · Cloudinary · Google Gemini
• DevOps & Tools: Docker · GitHub Actions (CI/CD) · Git · Vercel · Railway · Postman · Figma · Linux
• Testing: Vitest · React Testing Library

Also worked with: PostgreSQL · Prisma · Python · Firebase · Supabase