Alikamatu Osama

End-to-end ownership—product, design system, full-stack Next.js, AI integration with Anthropic Claude, Paystack subscriptions, Cloudinary uploads, VAPID web push, service worker, and production docs.

• ▸Local-first architecture—Zustand with persist middleware as single source of truth across pages. Every mutation runs synchronously; no network round-trips block the UI. The action API is shaped so a future DB swap (MongoDB / Postgres) replaces only the action bodies, not component contracts.
• ▸Custom design system instead of a UI library—five themes via CSS variables (OKLCH), zero borders / blur / glass / colored shadows. Includes hand-built DatePicker, TimePicker, Select, Modal, Popover, and a ~80 LOC markdown renderer to keep the bundle tight.
• ▸Streaming AI with classified errors—every AI surface (chat, plan my day, goal diagnose, task breakdown, note assistant) streams Claude responses via ReadableStream. Errors are mapped to typed kinds (no_credit / rate_limit / auth / overloaded) and rendered as markdown inside the same stream, so the UI never has to handle a separate error path.
• ▸Keyboard-first interaction—⌘K command palette with fuzzy matching, chord nav (g t for Today, g j for journal), quick add (N), cheatsheet (?), and per-modal focus traps. Every native form control replaced with custom primitives so the experience is identical across browsers and themes.
• ▸Paystack billing in GHS—Pro at GH₵70/mo or GH₵640/yr (saves GH₵200). HMAC SHA512 webhook signature verification with timing-safe compare; invoice list pulled live from the Paystack API.

Challenge: A local-first store powered by Zustand persist creates a server/client hydration mismatch on every authed page: the server has no localStorage, so it renders the seed; the client rehydrates and renders the user's real data. React flags this as a hydration error and discards the tree.

Solution: Built a 30-line Hydrated wrapper at the layout level that renders a structural skeleton on both the server and the first client paint, then swaps to children after the store rehydrates. No mismatch, no flash, one render. The same primitive removes a whole class of bugs across every store-backed page.

Pre-launch beta—actively used for personal daily planning. Service worker initially intercepted /api/* during PWA install, briefly breaking the streaming AI responses. Fixed by adding an explicit URL path check in the SW fetch handler and version-prefixing every cache so old caches purge cleanly on each new deploy.

Wire MongoDB for cross-device sync and persistent push subscriptions. Ship Google Calendar / ICS sync. Add a Vercel Cron job for goal-at-risk nudges (the endpoint is already wired). Add MFA. Move soft-delete retention enforcement from client-side to a server-side cron once the DB lands.

End-to-end ownership—auth, payments, admin dashboard, APIs, and a four-app architecture: Student App, Hostel Admin App, Super Admin App, Central Backend API.

• ▸Chose PostgreSQL for transactional consistency across multi-role bookings.
• ▸Used server-side rendering for SEO-optimized hostel discovery pages.
• ▸Centralized auth logic in NestJS backend to reduce duplication across all four apps.

Challenge: Handling concurrent bookings without double reservations when multiple students book the same hostel simultaneously.

Solution: Implemented database transactions with row-level locking (SELECT FOR UPDATE) and optimistic concurrency control to prevent race conditions.

A deployment caused auth tokens to invalidate—fixed by versioning JWT secrets and adding rollback steps to the deployment pipeline.

Add background jobs for email notifications and move heavy geolocation calculations off the request cycle using queue workers.

Full monorepo ownership: NestJS 11 API (with admin module), Next.js 16 buyer + entrepreneur storefront, shared packages, Paystack split-payment ledger, Google OAuth + TOTP/SMS 2FA, seller verification workflow, bulk CSV product import, and full SEO surface (sitemaps, robots, OG, JSON-LD).

• ▸Designed role-based access (Admin, Seller, User) with JWT, Google OAuth, and 2FA (TOTP + SMS) from day one to avoid retrofitting auth later.
• ▸Built a double-entry financial ledger over Paystack subaccounts and split payments so every payout is reconcilable end-to-end.
• ▸Turborepo + pnpm workspaces with shared @vendly/types so the API and storefront stay in lockstep on contracts.
• ▸Cloudinary for image + video pipelines—product video previews autoplay muted to boost discovery engagement.
• ▸Redis cache with in-memory fallback so dev and pre-production environments work without infra dependencies.

Challenge: Reconciling marketplace payouts across split payments, refunds, and return requests without drifting from Paystack's source of truth.

Solution: Implemented a double-entry ledger keyed off Paystack subaccount IDs with idempotent webhook handlers and a return-request workflow that posts compensating entries instead of mutating originals—every balance is reproducible from the journal.

Pre-launch—currently in final deployment preparation. Redis connection issue resolved locally during development.

Launch mobile app (Phase 2) after validating web platform with first 20 sellers and 100 orders; add escrow holding period and dispute resolution UI on top of the existing ledger.

Designed and built the complete website—service pages, booking flow, fleet showcase, ADA compliance, and brand execution.

• ▸Built to ADA accessibility standards from the ground up—not retrofitted.
• ▸Structured service pages for SEO around medical transport search terms.
• ▸24/7 dispatch integration and clear CTAs optimized for mobile users who need urgent transport.

Challenge: Presenting a regulated medical service in a way that builds immediate trust for users in vulnerable situations.

Solution: Led with safety certifications, ADA compliance badges, and driver credential information above the fold—reducing friction for users who need to book quickly.

Mobile responsiveness issues on older iOS devices with the fleet showcase—fixed with targeted CSS overrides and testing on Safari.

Add online booking/scheduling system with driver availability and SMS confirmations.

Full-stack ownership—bid ingestion system, CSV/Excel processing, real-time analytics dashboard, REST APIs, and audit logging.

• ▸Chose PostgreSQL for ACID compliance to prevent duplicate bid entries.
• ▸Used server-side CSV parsing to handle large files without blocking the UI.
• ▸Centralized bid validation logic to ensure data integrity across concurrent uploads.

Challenge: Preventing duplicate bids when processing thousands of rows from multiple concurrent file uploads.

Solution: Implemented database unique constraints with composite keys and transaction-based batch inserts with conflict resolution.

Large CSV uploads caused memory spikes and timeouts—fixed by implementing streaming file processing and chunked database inserts.

Add background job processing for large file uploads and implement Redis caching for frequently accessed bid analytics.