Sharvarok

Project Overview

Sharvarok did not start as a website. Its classifieds began as a section in the printed newspaper of Hromada, the Ukrainian-American non-profit whose platform I also build - the same PDF issues that are now in that site's archive. People were buying, selling and hiring through a page in a monthly paper, which works until it doesn't: an ad is out of date the day it is printed, nobody can search it, and the audience is whoever happens to hold a copy. So the section came off the page and became its own product: a searchable bilingual marketplace, and then iOS and Android apps on the same backend. Today it is a bridge between Ukrainians in the U.S. and local businesses, service providers and individuals - with the moderation, payments and tax handling a real marketplace needs and a newspaper page never did.

Challenges

• Building a scalable classifieds platform capable of handling a high volume of listings and user interactions.
• Ensuring fast and accurate search and filtering across diverse listing categories.
• Providing a secure messaging system while maintaining user privacy.
• Optimizing performance to handle high traffic and real-time listing updates.
• Enabling automatic English-Ukrainian and Ukrainian-English translations for listings.
• Getting people to post at all: the audience is bilingual, and a long form in a second language is the point where most of them give up.
• Charging the correct US sales tax per jurisdiction, which is a legal obligation rather than a rounding detail.
• Keeping a mobile app in behavioural parity with the website when both run against the same backend and diverge the moment one gets a feature first.
• Preserving traffic attribution across an OAuth round trip, which otherwise discards the UTM parameters and referrer that say where a user came from.
• Answering questions about traffic and listings without running heavy aggregate queries over the raw page-view table on every dashboard load.
• Surviving vulnerability scans against a public marketplace, and knowing what to do when one turns into a real incident.

Solutions

• Developed a modular, scalable architecture using Next.js and React for a fast, SEO-optimized front-end.
• Implemented Sphinx Full-Text Search for fast and accurate classified listing searches.
• Integrated Stripe and PayPal APIs for secure payment transactions when needed.
• Added a secured messaging system to allow private communication between buyers and sellers while ensuring data privacy.
• Leveraged Next-Intl for automatic English-Ukrainian and Ukrainian-English translations in listings.
• Deployed the platform on AWS EC2 with AWS S3 & CloudFront for media storage and optimized content delivery.
• Automated CI/CD workflows with GitHub Actions to streamline updates and maintenance.
• Implemented robust monitoring with Sentry and Google Analytics to track performance and user behavior.
• Built a React Native client for iOS and Android on the same Next.js API as the website, with Stripe PaymentSheet for in-app payment, Firebase messaging and analytics, and Sentry, released through scripted App Store Connect and Google Play pipelines that are dry-run by default.
• Added voice dictation to the posting form that feeds an AI parser: the seller describes the item aloud in Ukrainian or English and the listing fields come back filled.
• Implemented US sales tax as its own subsystem with a documented test plan, rather than a rate constant buried in the checkout.
• Built DB-backed feature flags with an admin UI and a short cache, so a half-finished feature can ship dark and be switched on without a deploy.
• Added MaxMind GeoIP through a shared reader that re-reads on file change, refreshed by a weekly cron, so location-aware listings do not depend on a third-party lookup per request.
• Preserved UTM parameters and referrer across the OAuth round trip, so sign-ups keep the attribution that says which channel produced them.
• Rolled page views into daily aggregate slices on a fifteen-minute cron, so the analytics dashboard answers from a summary table instead of scanning raw events.
• Generated Open Graph preview images for business profiles at the size and thresholds Facebook actually honours, with a backfill for existing records.
• Built bilingual structured fields for jobs and resumes - the source plus paired translated fields - so a vacancy is searchable and readable in both languages rather than machine-translated at display time.
• Added a health endpoint that distinguishes fatal from degraded - the database is fatal, Redis and Sphinx are degradation - plus a post-deploy smoke run over public URLs that bypasses the CDN cache and reports to Telegram.
• Ran end-to-end tests in Cypress alongside the smoke suite, and documented a real vulnerability-scanning incident with the log analysis and the reporting path in the US.

Key Features

• Multi-Category Listings - Users can post and browse ads for services, jobs, real estate, and more.
• Advanced Search & Filtering - Full-text search powered by Sphinx for fast and accurate results.
• Secure Transactions - Integrated Stripe and PayPal for seamless payments when applicable.
• AI-Powered Translations - Automatic English-Ukrainian and Ukrainian-English translation for listings using Next-Intl.
• Secured Messaging - Private messaging system with a focus on data privacy and security.
• Optimized Performance - Hosted on AWS EC2 with AWS S3 & CloudFront for efficient content delivery.
• CI/CD & Monitoring - GitHub Actions for automated deployments, Sentry & Google Analytics for tracking performance.
• Native Mobile Apps - A React Native client on iOS and Android against the same backend as the website, with in-app Stripe payment and push notifications.
• Speak Your Listing - Voice dictation in Ukrainian or English parsed by AI into filled listing fields.
• US Sales Tax - Computed per jurisdiction as its own tested subsystem.
• Feature Flags - Database-backed flags with an admin UI, so features ship dark and switch on without a deploy.
• Bilingual Jobs & Resumes - Structured fields carried in both languages rather than translated at display time.
• Analytics Dashboard - Daily page-view rollups on a fifteen-minute cron, with traffic attribution preserved across OAuth.
• Health Checks & Smoke Tests - Fatal versus degraded distinguished explicitly, with a post-deploy run reporting to Telegram.