RocketVerify — Custom Async SMTP
Designed and built a scalable email-verification service implementing a custom async SMTP protocol, with Playwright fallback for the hardest cases. Throughput at ~500M emails/month.
Replaced an off-the-shelf vendor stack with an in-house service that gave us throughput and economics no vendor would.
Problem
Email verification is the central, most expensive step in our core product. Vendor pricing scaled poorly with our volume, throughput hit hard ceilings, and edge cases — catch-all detection, soft bounces, providers that adaptively blacklist, the long tail of weird MTAs — were either not handled or not handled the way we needed.
Design
A purpose-built async SMTP client that owns the connection lifecycle: domain-aware connection pooling, MX rotation, per-domain rate limits to avoid IP-reputation damage, retry policy that distinguishes between greylisting, soft bounces, hard bounces, and adaptive blacklisting, and end-to-end observability per verification.
For domains that resist standard SMTP probing, a Playwright fallback path performs heavier-weight checks — pooled, throttled, and cost-bounded so the long tail doesn’t dominate the budget. The protocol layer was authored against a careful read of the relevant RFCs, not a wrapped library.
Outcome
Throughput scaled to roughly 500M emails per month. Beyond raw scale, the service became the substrate that unblocked two further wins:
- Offline verification — ML-prioritized batch verification that cached the most expensive part of lookups about 90% of the time.
- Lookup parallelization — verification became the parallel hot path inside the lookup, dropping latency from ~12s P90 to ~3s P90.
Together with the 2025 prediction rewrite, lookup success rate moved from 55% in early 2024 to ~90% by end of 2025 — a direct credit-multiplier on the company’s revenue.