Ghanshyam Khokhariya

VisionLab is a live AI platform I built and launched single-handedly in two months. Product photography is slow and expensive — a seller waits weeks and pays studio rates for images. On VisionLab, they upload one phone photo and get studio-grade shots, lifestyle scenes, posters, and video creatives in under a minute, with an ad engine that connects Meta and Google accounts and turns the best creatives into campaigns. Under the surface, every generation is routed to the cheapest AI model that can satisfy it — across FLUX.2, Veo, Kling, Ideogram, and more, with Claude classifying images and writing copy, and every credit is protected by a double-entry ledger on PostgreSQL, so a failed or crashed job can never charge anyone twice. Payments are processed via Razorpay subscriptions with verified webhooks, and every generated asset ships with content credentials and watermarking. I designed, built, deployed, and operated all of it — front end, back end, database, workers, and the underlying cloud infrastructure.

Star Health is a React Native healthcare platform for Star Health and Allied Insurance, one of India's largest publicly listed insurance companies. The app is live with 5M+ downloads and 300K+ Play Store reviews. I built the doctor discovery system with real-time multi-filter search across tens of thousands of profiles using debounced multi-stage fetching with local cache reconciliation, and the telemedicine chat system with client-side message queuing and platform-specific reconnection strategies for iOS and Android. I diagnosed a Hermes-specific memory leak that only appeared after 45+ minutes of Android use — invisible to JavaScript profiling — traced to native bitmap references in the bridge layer using Android Studio's native heap profiling.

OnSPARK is Ontario’s sector-governed learning health system for long-term care, with 200+ participating homes and 6 billion data points processed, powering near-real-time insights for quality improvement, research, and public health surveillance, in partnership with institutions including McMaster University. I built the front end across both of its applications in Next.js, React, and TypeScript as the sole front-end engineer: an authenticated multi-tenant admin dashboard where organization staff manage users, work a notifications inbox with email reply, and analyze resident health metrics through heatmap, trend, and correlation charts — with resident-level views scoped to the selected facility — and the bilingual public website with full English and French coverage. The engineering that mattered: virtualizing the organization and facility switcher so it stays smooth with hundreds of entries, a unified conditional-fetching pattern so charts degrade to explicit empty states instead of hanging when the back end is down, and response validation on login so a malformed payload can’t take down the dashboard.

Lungta is the mobile app where customers can order a courier without creating an account: set pickup and delivery, describe the merchandise, choose between scheduled delivery and urgency tiers priced as a live quotation, verify contacts by SMS or email security codes, pay through Stripe with Apple Pay or Google Pay, and track the courier in real time on a map.

I led the React Native build of the client application from first commit to production, in TypeScript. I architected the app, turned the product designer's Figma system into a Storybook-documented component library, and built the guest ordering flow end to end — the quotation-expiry system that returns customers to updated pricing when a quote lapses, payment decline, and cancellation handled as designed states, live tracking over WebSockets, and notifications with synchronized badge counts and deep links. Couriers work their routes through the same application via a gated entry I built, which also includes proof-of-delivery components, including the signature-capture canvas. Architecture documentation and deployment runbooks are shipped alongside the codebase.

StockConnect is a dropshipping platform by Olltek built for the Saudi market — vendors based outside KSA sell into regional marketplaces like Noon and Namshi from one place, managing their catalog once instead of per marketplace, across seller brands including Metamorphoza, Cherry & Jerry, and Respire. Vendors onboard catalogs by spreadsheet, Unicommerce ERP integration, or Shopify — connected via API, OAuth, or secret key. Publishing runs through a single mediator integration with ChannelEngine, which holds the marketplace configurations: a vendor uploads a product once, and it publishes to every connected marketplace, with a category mapping module bridging each marketplace's taxonomy and webhooks routing order and shipping events to the seller they belong to. The platform is multi-tenant with a three-role hierarchy — Master Admin across all brands, Super Admin scoped to a seller, Admin below — with commercial data gated by role. I built the live inventory dashboard at its center: thousands of SKUs updating via a live feed with free filtering; filter state in URL params as the single source of truth; memoized filtering isolated from the update stream; and virtualized rendering so live updates touch only the cells that changed.

I built internal commerce tooling for Styli, a fashion and lifestyle eCommerce platform backed by Landmark Group, one of the Middle East’s largest retail conglomerates, operating across Saudi Arabia, UAE, Kuwait, Bahrain, Oman, and Qatar with 400+ brands. I developed a bulk order processing system allowing buyers to upload Excel files containing product details, including sizes, colors, types, and photos, with automated seller-side processing and dispatch workflows. I built the Styli companion app integration, enabling sellers to upload products and automatically synchronize them in real time to the main platform. I developed seller dashboards covering inventory management, order processing, shipment tracking, and packet labeling, and optimized front-end performance and application state management using Redux under high-traffic eCommerce load.

Ollkom is the customer-facing portal for Alhome, Saudi Arabia's online furniture and home decor store. I built the full user experience in Next.js with SSR, implementing Auth.js for secure authentication, product listing across collections with advanced filtering, wishlist, cart, and checkout with integrated payment support, including Tabby and Tamara installment payment options. Localization was built using next-intl, supporting English and Arabic with full RTL compatibility. User account features covered order history, account details, and policy management, with a responsive design optimized across desktop and mobile.

Olltek Commerce Admin Portal is the back-office management system for Alhome, a Saudi Arabian furniture and home decor eCommerce platform. I built the full admin portal in Next.js with SSR, covering catalog management across products, variations, collections, attributes, and customer reviews. I implemented order management workflows, including tracking, status updates, and shipment processing, alongside customer onboarding and payment handling. Marketing tooling covered coupon creation, bulk coupon uploads, discount configurations, and an influencer and affiliate campaign management module. I applied RBAC across all modules for permission-based access control and built multi-language support using next-intl for international back-office users.

OpenTalent is an AI career copilot platform with a network of over 40,000 engineer profiles. I built the recruiter-facing admin portal, where accessibility was written into the acceptance criteria: every flow is completable by keyboard with real focus management — modals that trap focus and return it on close — semantic structure with native elements, form errors wired via aria-describedby and announced, WCAG AA contrast with color never carrying meaning alone, and testing enforced through axe and jest-axe in CI plus manual VoiceOver passes. The profile directory is server-side filtered and paginated — 40,000 records don't ship to a browser — with the filter panel serializing into API queries and skeleton rows holding layout during fetch. I also built the conversational assistant interface: an append-only message log with stable IDs reconciling optimistic sends against server confirmations, token-by-token streaming replies, scroll pinned only when the user is at the bottom, and the input gated on an explicit idle–sending–streaming–error state machine so send, stop, and retry never contradict each other.

BatchLeads is a React Native real estate platform trusted by 10,000+ professionals across the US for lead generation, property search, and virtual canvassing. I built high-density map rendering by bypassing React's reconciliation entirely — moving property markers to native GeoJSON overlays, which are passed directly to the native map SDK via a custom bridge module, eliminating JavaScript thread freezes at scale. I resolved a GraphQL subscription cascade that triggered full component tree re-renders across 10,000+ properties, fixing it with normalized Apollo cache updates and viewport-aware render prevention. I also built a virtual canvassing tool with real-time driving route tracking and implemented a multi-filter property search across 300+ data filters.

Everty Driver is a React Native EV charging platform for Everty Pty Ltd's Australian charging network, live on the App Store and Google Play. I implemented real-time charging session management using Socket.IO — including session start and stop, live battery status monitoring, and real-time rate updates synchronized between the driver app and the operator dashboard. I resolved concurrent socket state conflicts between the two clients by implementing optimistic concurrency control with version tokens, triggering full state re-fetch on conflicting updates rather than partial merges. I built the real-time charger map using Google Maps, integrated credit card payments with pre-auth prompts and idle-fee support, added RFID card management with NFC scanning, and managed multi-environment Expo builds across development, staging, and production.

I worked on the Time To Be Generous React Native Expo application during its final development phase, focusing on stabilization, bug resolution, and preparing it for iOS App Store release. I implemented iOS native offer code redemption for subscription flows, resolving platform-specific billing constraints within the Expo architecture. I handled the full iOS release pipeline, including bundle ID configuration, App Store Connect setup, EAS builds, and TestFlight distribution. I fixed UI bugs identified during QA, implemented homepage and navigation updates, and resolved localization inconsistencies between English and French content throughout the app.

ECHO is a cross-platform social storytelling app — React Native and TypeScript on a Supabase back end, in beta on iOS and Android. It takes the familiar Instagram-style story format and adds something incumbents don't have: "Chains" — stories multiple users co-author in sequence, including private, host-gated communities. Users post photos, videos, text, and voice-over stories.

• Architected the Supabase back end around Postgres RPC functions and triggers, keeping business logic in the data layer and the client thin — so story, chain, and chat state stay authoritative as multiple users act concurrently.
• Engineered the "Chains" feature: sequential multi-contributor stories with request-to-join access and dual gating through membership and host approval.
• Built real-time messaging on Supabase Realtime — reactions, typing indicators, and read receipts — designed to stay consistent as delivery and presence change live.
• Implemented an offline-first action queue and local chat cache so user actions persist and replay cleanly on reconnect instead of failing under poor connectivity.
• Built the native media pipeline — Vision Camera, video with thumbnails, and voice-over audio — behind a unified multi-format story creator.