Building WordPress products, SaaS platforms, and developer tooling — at scale.

Problem. A major version bump required coordinating breaking changes, a new brand color, redesigned onboarding, refactored emails, a React-based admin/vendor panel, and full backward-compatible commission calculation — without breaking sites already live in production.

What I did.

• Defined the release tracker, version-bump strategy, and changelog.
• Drove the V4 manual-testing program covering Withdraw, Coupon, RMA, Staff, Store SEO, Shipping, Vendor & Product Subscription, Announcement, Admin/Vendor Analytics, Admin Onboarding, Manual Order Creation, Commission backward-compatibility, Dokan Intelligence AI, and the new brand color.
• Maintained the V4 bug tracker and resolved release-blockers across documentation, modals, dummy-data imports, and help-center URLs.
• Removed deprecated code paths ahead of the cut.

Impact. Shipped V4.0 to production with a coordinated test matrix across 18+ modules. The release-engineering pattern I established (sub-issues per module, "Awaiting QA" labels, parent tracker) is now reused for every minor release — including Dokan 4.2.0.

Problem. Dokan's vendor and admin panels were a tangle of jQuery, Vue, and templated PHP — global mutation, inconsistent state management, slow module rollout, hard to test.

What I did.

• Built the Dokan Data Store and a React HOC library — the foundation every subsequent module reused.
• Established API development guidelines and frontend slot-fill / filter standards so other engineers could ship consistently.
• Migrated five vendor-panel modules end-to-end — Announcement, Store SEO, Product Subscription, Shipping, Staff — plus Vendor Dashboard Withdraw.
• Shipped the React-based Admin Panel Header and unified admin dashboard structure.
• Removed global mutation by routing everything through Dokan Modal.

Impact. The vendor panel went from "every module its own jQuery snowflake" to a consistent React app with a shared data layer. Module rollout time dropped meaningfully and the patterns set here are still being followed.

Problem. The admin dashboard was slow and the analytics reports were wrong. WooCommerce's analytics tables (wc_order_stats, wc_order_product_lookup) double-counted Dokan suborders and their refunds, inflating GMV. Marketplace operators couldn't trust the numbers.

What I did.

• Shipped optimized queries & APIs for the new admin dashboard: Monthly Overview, TODO Section, Customer Metrics, All-Time Marketplace Stats, Top Performing Vendor, Most Reviewed Product.
• Built the implementation plan, component layer, and WC Admin Analytics ↔ Dokan compatibility.
• Authored a seller-analytics RFC proposing a new dokan_order_stats table that mirrors WC's schema but adds seller_id and is_suborder — cleanly separating seller earning, gateway fees, discounts, admin commission, and admin subsidy (when an admin-funded discount exceeds the commission, e.g. a $20 discount on a $100 product at 10% commission means the marketplace eats $10).

Impact. Admin dashboard load improved; the operator-facing "all-time marketplace stats" widget exists today because of this work. The seller-analytics RFC sets up the next phase of fixing analytics correctness at the data-warehouse layer.

Problem. Marketplace payment gateways aren't just "process the card." You're splitting funds between admin and vendors, handling per-vendor onboarding, supporting refunds (which redistribute commission backward), and surviving webhook retries that must be idempotent.

What I did — Paystack (shipped end-to-end).

• Vendor onboarding via Paystack sub-accounts.
• Make-payment with multi-party split.
• Refund logic that correctly reverses commission.
• Webhook handler for async settlement events.

Stripe Express. Integration with WC Stripe; refactored emails for Stripe / Stripe Express; resolved guest-checkout PHP warnings.

Vendor Subscription module RFC (open). Designed the next-gen module covering switching (upgrade/downgrade with returnable amounts), renewal with grace periods, cancellation with trial-period handling, and extension offers — including a worked example of grace-period overlap when one subscription fails and the vendor subscribes to a different plan.

Impact. Paystack is in production powering African-market vendors. The subscription RFC codifies edge cases nobody had written down — switching, grace periods, cancellation timing — into a single design that handles credit cards as a first-class payment method.

Problem. The Dokan mobile app needs push notifications driven by WordPress events (new order, low stock, refund, announcement) — delivered in the user's preferred language across 100K+ devices, with reliable delivery and no synchronous coupling to the WP request lifecycle.

What I did. Authored a full feature spec broken into 6 implementation sub-issues so the team could parallelize:

• WordPress Plugin Core — hook integration & OneSignal API client.
• Multilingual content handling via WPML — translations resolved server-side.
• User segmentation & targeting — role, location, content-type filters.
• Flutter app integration — OneSignal SDK, lifecycle, deep linking.
• Notification management & admin UI.
• Non-functional requirements.

Notable design decisions.

• End-to-end latency target < 10s (WP hook → device), with < 500ms WP-side budget.
• No synchronous OneSignal calls from WP hooks — everything goes through Action Scheduler, so request threads stay fast and notifications retry independently.
• Retry policy: exponential backoff at 1s / 4s / 16s (3 retries), then dead-letter logging.
• Scale strategy: at 100K+ users, fan-out happens inside OneSignal segments — never iterate users in PHP.
• Security: REST API key in a wp-config.php constant (not wp_options), HTTPS only, XSS-sanitized payloads, manage_options capability checks, rate-limited manual sends.

Impact. This is the design document the implementation team is now building from. The decomposition (6 parallelizable sub-issues with explicit interfaces) is the same pattern I use for every multi-engineer feature.

A reusable React + TypeScript component library that powers schema-driven settings UIs across multiple WordPress plugins. Plugin teams ship admin UIs without writing UI code; the library bundles shadcn sidebar primitives, a WP Media upload field, and other WP-specific bridges.

WordPress SMS notification plugin. Deep refactor: 21 new extensibility hooks, untangled SOLID violations between gateway/formatter/dispatcher, modernized the Node/PHP toolchain, and migrated the settings UI to plugin-ui.

An MCP server that lets AI agents manage WordPress sites by speaking the official WordPress/abilities-api (209★). Contributed upstream while building the consumer; passed the WordPress.org Plugin Check with zero errors and zero warnings.

An open-source Laravel Scout engine for Azure AI Search — drop-in driver enabling Laravel apps to power full-text search via Azure. Implements the Scout engine contract end-to-end: indexing, search, pagination, filters.

A Flutter mobile app for Enzaime's Pharmacy Management System — auth, point-of-sale, inventory, reports — built end-to-end alongside the PHP backend. Designed the offline-first data model so POS workflows survive intermittent connectivity, and set the architectural pattern (state management, repository layer) for the wider Enzaime mobile portfolio.

A personal resource planner SaaS where I own the full stack — API, web, and auth services — across 9 repos. The auth-api / api-app / web split lets each service scale and deploy independently; tips-api is the highest-velocity codebase in my portfolio (118 merged PRs).

Bangladesh's first complete health-assistant app — selected for the Slush Event 2016 from Bangladesh. Owned database design, architecture, coding, testing, and deployment. Stack: Laravel + jQuery + VueJS + MySQL + Memcached + Redis + IonicJS.

Bangladesh's premium online travel service. Platform upgrades, module refactors, and new-feature delivery during a fast-growth period. Stack: Laravel (PHP), jQuery, MySQL, Couchbase, AWS.