AI Newsletter Delivery & Operations Platform

Overview

A six-author production email platform I contributed to between August 2024 and May 2026, authoring 70 of its 1,068 commits and pairing with a coworker from July 2025. I did not build the sending engine, and this page does not claim otherwise. My scope was ad-network and brand integration, configuration of the email service provider send consumers, and segmentation for scheduled sends. In practice that meant onboarding new advertisers and brands through the ad entity and the email service, wiring and configuring the Postmark, Mailgun and Resend consumers — including the Resend migration — and building the recipient segmentation that decides who a scheduled provider send goes to. The files I touched most were the Postmark and Mailgun send consumers, the email service, the cron scheduler and the webhook controller. Two commits went to the OpenAI service that rewrites subject lines to reduce spam scoring, and one maintenance script restores ad titles and calls to action from CSV. TODO(razvan): decide the public name for this project. content-spec.md D4 leaves it open between "WellnessInbox" and "yt-newsletter", and the title in the frontmatter above is a neutral placeholder, not a decision. Blocked on D3 alongside the disclosure question. TODO(razvan): confirm the framework. The file naming strongly suggests NestJS, but OWNERSHIP.md only establishes TypeScript and the three providers. Do not list a stack you have not checked.

Third Party Integrations

TODO(razvan): this is the section where a contribution case study is strongest, because provider integration is literally what you did. One H3 per provider, each a bulleted list of what you configured and what went wrong. Write it from the commits, not from memory.

Postmark

TODO(razvan): what the send consumer does, what per-brand configuration it takes, what the webhooks feed back into.

Mailgun

TODO(razvan): same shape as above, plus what differs from Postmark and why both exist.

Resend

TODO(razvan): the migration. What moved, what broke, what the fix was.

Ad networks and brands

TODO(razvan): the integration shape — what an ad network needs in order to be onboarded, and what a brand adds on top of it. Name nothing until D3 is resolved; describe the shape instead.

Challenges and Learnings

TODO(razvan): one H3 per challenge, prose underneath. Two or three is enough for a contribution entry. Keep the frame honest — "here is a specific problem I owned inside a larger system" is a stronger story than an implied claim on the whole platform.

TODO(razvan): running three email providers at once

TODO(razvan): why a second and then a third provider existed, and what configuration had to be duplicated across consumers.

TODO(razvan): segmenting a scheduled send

TODO(razvan): how recipients get selected, and what made it harder than a WHERE clause.

TODO(razvan): contributing to a codebase you did not design

TODO(razvan): optional, and only if you want it. The honest version of this — how you got useful inside a 1,000-commit codebase written mostly by someone else — is a genuinely good interview answer. It is also the most senior thing on this page.

Outcome

TODO(razvan): no outcome numbers are established for this project, and the ones that would matter most — send volume, deliverability change, revenue processed — are the ones most likely to be confidential. Resolve D3 first. TODO(razvan): if the numbers cannot be published, describe the outcome in shape rather than magnitude, and be careful to keep it scoped to your contribution rather than the platform's.