Qelv

Back to Work

Internal product

A self-hosted email verification engine built for explainable results

An internal product: a bulk verification CLI with DNS and multi-stage SMTP checks that explains each classification instead of emitting a black-box score, deployed on its own verification node and run on real lead lists.

Internal product · Delivered · Email infrastructure

Source not public

Our own engineering. We will screen-share the code, the architecture, and the tests with anyone seriously evaluating us; publishing the repository invites the scrapers we spend our days defeating.

Request a code walkthrough

Headline figure
34 reason codes · 25 output columns

documented classification system; every verdict traceable to the check that produced it, June 2026

Status
Delivered

Internal product

Practice
Build · Data

Email infrastructure

Sources
2 figures

engineering record

The problem.

Bulk email verification services return a single opaque score at a price. Operating our own cold-email outbound meant needing something cheaper that also explained itself: why an address was classified the way it was, especially on catch-all domains, which accept mail for any address and so defeat a plain mailbox check.

What was built.

  • Python 3.10+ bulk verification CLI using DNS and multi-stage SMTP checks
  • Provider profiles with greylisting and tarpit handling
  • Catch-all domains given a score rather than passed or rejected outright
  • SPF, DMARC, PTR, TLS, legitimacy, confidence, and bounce-risk signals
  • 34 documented reason codes and 25 output columns per address
  • Concurrent bulk runs via multiprocessing
  • pytest suite and GitHub Actions workflows for CI, CodeQL, deployment, and health
  • Deployed on a dedicated verification node and run on real lead lists, with results delivered as CSVs

Timeline.

  1. 3 March 2026

    Research session: the target set as matching a commercial tool's reason codes on a 99,000-address benchmark set, not just its verdicts; catch-all domains to be scored rather than passed or rejected; SMTP-probing prototype committed

  2. 6 March 2026

    Repository created; provider profiles, 34 reason codes, 25 output columns, a pytest suite and GitHub Actions workflows for CI, CodeQL, deploy and health landed over the following three months

  3. 24 June 2026

    Production run from the verification node on a 602-address lead list: 421 valid, 11 invalid, 157 risky, 13 unknown in 742 seconds, each with its reason code; results delivered as four CSVs

  4. 27 June 2026

    Next version of the verifier started in a second repository

What this case claims, and what it does not.

On the record

  • Architecture and operating maturity, as built
  • Internal product, clearly labeled as such
  • The June 2026 run summary, as recorded

Withheld, and why

Figures that exist internally and are not on this page, with the reason.

An accuracy percentage
There is no published benchmark methodology yet, so no accuracy figure is claimed and no comparison to a vendor is made.
Where it runs
Production hostnames, sending identities, and addresses stay private.

Where to next

Have a problem in this shape?

Send a short description. The form opens with this case already named, so you only add the part that is yours. We will tell you whether it is worth building at all.

We reply within two business days. If it is a fit, you get a one-page note on what to do first, before any sales call.