SYS JS.DEV
BUILD F3CREL
DATE 2026.04.26
UTC 01:30 UTC
LOC NYC → STANFORD
STATUS OPEN TO ML/SYSTEMS ROLES

Email Verification + Domain-Success Prediction

Rewrote in-lookup email verification to run in parallel and prioritize the right candidates first, plus a full AI domain-success tracking system. Doubled prediction success and drove lookup success from ~55% to ~90%.

Lookup success 55% → 90%
Prediction success
Span BEGINNING-2024 → END-2025

The single largest revenue lever I’ve moved at RocketReach.

Problem

Email verification is the most expensive and most failure-prone step in the lookup pipeline, and the lookup pipeline is the core product. Two compounding problems by 2024:

  1. Verification ran sequentially on a candidate list, so the lookup waited on the slowest probe.
  2. Domain-success prediction — given an unknown email at a known domain, will it land? — was based on legacy heuristics that aged badly as the email-deliverability landscape shifted.

Both ate directly into the lookup-success rate, which is the lever for credit consumption and revenue.

Design

The verification rewrite parallelized probes across candidates and added an explicit prioritization that ran the candidates most likely to succeed first, short-circuiting the rest of the wave the moment one landed. Throughput accounting moved into the call path itself so backpressure and cost were continuously controlled.

The domain-success prediction layer was rebuilt as an AI-driven tracking system that ingests outcomes per domain, models drift, and re-prioritizes candidate strategies as deliverability shifts. This piece sits behind every email lookup the product makes.

Outcome

STACK · Python · AWS · LLMs