Verifox.ai

Email lists decay fast — people change jobs, domains die, and typos slip in. Sending to those addresses burns money, tanks sender reputation, and quietly kills deliverability. Most verifiers either guess from syntax alone or are too slow to sit inside a live signup flow. Teams needed verification they could trust in real time and at bulk scale.

1. 01

   A 9-point engine with a Rust core

   Syntax, MX lookups, live SMTP conversations, IMAP and mail-auth checks, and disposable-domain detection run cheapest-first and short-circuit the moment a verdict is certain. The hot path lives in a dedicated Rust validation service so it stays fast under load, fronted by a Hono/TypeScript API.

2. 02

   AI catch-all resolution

   Catch-all domains accept everything at the SMTP layer, so traditional checks return 'unknown'. We resolve them with an AI pass over historical signals — embeddings stored in a Qdrant vector DB plus graph clustering (graphology / Louvain) — backed by Anthropic, OpenAI, and Cohere models. That converts most unknowns into a defensible answer: the hardest 10% of the problem.

3. 03

   Sub-50ms API + bulk pipeline

   The real-time path responds in under 50ms so it sits inside a live signup form. A separate Redis + BullMQ pipeline ingests CSV/Excel files of millions of rows, with results streamed back and credit billing metered per check through Stripe.

4. 04

   Leading the build

   As project lead I scoped milestones, owned the architecture and code review across a 5-engineer team, and shipped end to end — from the Rust core and SMTP layer to the dashboard and billing.

98.7%

Verification accuracy

<50ms

Real-time API response

9-point

Verification engine

5-eng

Team I led

Rust · Hono · TypeScript · MongoDB · Redis · BullMQ · Qdrant · OpenAI · Anthropic