PRNow — press release distribution platform

The product

PRNow lets businesses, agencies and resellers publish a press release without talking to a salesperson. You write the release (or let the AI writer draft it), pick a plan and any premium outlets, and pay. The platform distributes it, collects the live placement links and sends a branded report.

My role

I’ve been the lead developer since August 2025. The project started as a generated homepage mock-up; I built everything behind it: accounts, orders, payments, distribution, reporting, support and the admin back office. I work directly with the owner on priorities and ship most changes as reviewed pull requests — around 545 so far.

What I built

• AI press release writer. Drafts grounded in live web-search references, with keyword data, SEO and “LLM-readiness” checks, and a fallback across several model providers so one outage doesn’t stop customers.
• Order pipeline. Per-package status tracking, refunds to a credit ledger, automatic cancel-and-refund for orders that stall, and a duplicate-title guard.
• Partner fulfilment automation. Orders are submitted to wholesale distribution partners automatically, tracked, and their placement links harvested back into the order.
• Branded reports. Web, PDF and Excel reports, including white-label report domains for agencies.
• Checkout. Cards, Apple Pay and Google Pay, saved cards and crypto, with idempotent order IDs so a retry can never charge twice.
• Support desk. Tickets from email, Telegram and partner threads land in one inbox. AI drafts replies from the knowledge base and past tickets; a person always reviews before anything is sent.
• Growth tools. Built-in email campaigns with A/B variants and paced sending, admin analytics, and about 17 free SEO and PR tools.
• The network around it. Orders also fan out, through HMAC-signed webhooks, to a five-site newsroom network I built, and a white-label storefront lets resellers run their own branded PR shop on top of the API.

Hard problems

• Moving off Vercel without downtime. I moved the app to a VPS behind Cloudflare, rebuilt the scheduled jobs as system timers, and fixed client-IP handling so a forged header can’t bypass rate limits.
• An admin page that shipped an estimated 80 MB. The release list loaded every field of every release. I slimmed the list payload and load details on demand.
• AI that fails gracefully. Provider errors are classified and retried with jitter, and customers see a plain message instead of a vendor error.