// Artifact · Technical Writing

The Applied AI Brief — A Published Industry Column

A monthly newsletter on where AI is actually earning its keep, and the serverless email system that delivers it — written, built, and operated end to end

Technical WritingAWS ServerlessEmail InfrastructureResponsible AIFull-Stack

Open full experience

// 01 · Introduction

Introduction

The Applied AI Brief is a monthly column I research, write, and publish myself, arguing a thesis I keep seeing confirmed at work: the constraint on enterprise AI stopped being model capability and became the layer around it — permission, accountability, and plumbing. Three pieces are live: a researched industry brief covering healthcare, software engineering, and financial services, and two field notes drawn from my own builds. The column is not only writing, though. Because a portfolio site is a static export with nowhere to store a subscriber, I designed and shipped the email platform underneath it: a serverless, compliance-first subscribe and delivery system running in its own AWS account.

// 02 · Description

Description

The reader-facing half is a statically exported Next.js column: post data lives in one typed source file that renders the web pages, an RSS feed carrying full post content, and the email HTML, so the three can never drift. Charts are hand-written inline SVG driven by CSS custom properties, legible in both light and dark themes, and re-rendered as plain HTML tables for email, where many clients strip SVG. The delivery half is a single Lambda behind a function URL with four routes — subscribe, confirm, unsubscribe, send — backed by two DynamoDB tables: subscribers keyed by address with a token index, and a sends table keyed by (issue, address) so re-running a send is idempotent by construction rather than by care. Consent is double opt-in: an address is stored as pending and only becomes active after the subscriber clicks a link carrying a 32-byte random token, compared in constant time. Every issue carries a one-click unsubscribe link, List-Unsubscribe and List-Unsubscribe-Post headers, and a postal address. SES bounce and complaint events flow through SNS back into the same function, which suppresses those addresses automatically. The site itself is served from a private S3 bucket through CloudFront with Origin Access Control, and GitHub Actions deploys it by assuming a scoped role through OIDC federation, so no AWS credential is stored anywhere.

// 03 · Objective

Objective

To practice the part of an AI/ML career that is not model work: choosing what is worth saying about a fast-moving field, defending it with dated primary sources, and shipping the infrastructure that puts it in front of readers — under a real constraint of near-zero operating cost and no shortcuts on consent or compliance.

// 04 · Process

Process

01Defined the editorial thesis and the audience — AI/ML hiring managers, engineering collaborators, and practitioners — before writing anything, and set editorial rules that outlive any single issue.

02Researched Issue 001 from primary sources (FDA clearance coverage, company announcements, published benchmarks, regulatory timelines), listing all ten with dates and never publishing a figure absent from a listed source.

03Wrote two field-notes essays from my own builds, using only numbers my own evaluation files produced.

04Built the column as a typed data source rendering web pages, RSS with full content, and email HTML from one place, with hand-drawn SVG charts theme-aware via CSS variables.

05Designed the data model for the failure modes rather than the happy path: token-indexed subscribers, and a sends table whose composite key makes duplicate delivery impossible.

06Implemented the API with consent and privacy as constraints — double opt-in, constant-time token comparison, identical responses whether or not an address exists, honeypot and timing bot filters, and no full email address written to any log.

07Provisioned a dedicated AWS member account via Organizations to isolate the workload, then SES with DKIM, a custom MAIL FROM domain, and DMARC on a sending subdomain to protect the root domain's reputation.

08Wired bounce and complaint events through SNS into automatic suppression, so list hygiene requires no human in the loop.

09Moved the site behind CloudFront with a private origin, and replaced manual deploys with GitHub Actions OIDC federation — scoped to one repository, one branch, and three actions.

10Wrote the runbook: publishing sequence, dry-run preview, and the editorial rules that keep future issues honest.

// 05 · Tools & Technologies

Tools & Technologies

Next.js 16 · TypeScript · Tailwind CSS 4 (static export)AWS Lambda (function URL) · DynamoDBAmazon SES (DKIM · custom MAIL FROM · DMARC) · SNSAmazon CloudFront · S3 with Origin Access ControlAWS Organizations · IAM · Route 53 · ACMGitHub Actions with OIDC federation (no stored credentials)Hand-written inline SVG data visualization

// 06 · Value Proposition

Value Proposition

Unique Value

Most portfolio newsletters are a paragraph and an embedded signup widget from someone else's platform. This one is both halves: the writing takes a position and cites dated primary sources for every number, and the delivery system underneath it is one I designed for the failure modes — duplicate sends made structurally impossible, consent proven rather than assumed, complaints suppressed automatically, and no long-lived credential anywhere in the deploy path. It also demonstrates a judgment I care about: the compliance requirements are not decoration bolted on at the end, they are the schema.

Relevance to My Audience

For hiring managers it shows an engineer who can carry something from a blank page to a running, compliant production system alone — and who can then explain it to a non-specialist, which is the harder half of most technical roles. For readers, it is a monthly filter over a noisy field, written by someone building in it rather than reporting on it. And for me it is a forcing function: publishing a dated claim with sources attached is the fastest way to find out whether I actually understand something.

// 07 · References