The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Reliable AI-generated email takes more than a detailed prompt. In an October 2026 interview, AlpacaRelay co-founder and CTO Alina Olefirenko describes a production approach built around constrained MJML components, layered quality checks, visual review and real test sends. The system is also moving toward recommendations based on business data, although its connection between campaign rationale and later business outcomes was still in development at the time of the interview.
Olefirenko’s account is a useful engineering case study in the gap between an email that looks plausible and one a business can trust enough to send. The company’s approach replaces a single large block of model-generated HTML with smaller, defined components, then checks structure, meaning and appearance separately. Its central lesson is that generation is only one part of production readiness.
These details reflect Olefirenko’s description in The AI Journal interview published 5 October 2026; the interview does not provide independent technical testing or audited campaign results.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Why prompt-only HTML was not enough
Email is unusually unforgiving of loosely controlled output. Clients support HTML and CSS differently, and layouts still rely heavily on tables. A model can produce markup that looks reasonable while leaving the sender with uncertainty about structure, rendering and whether the message fits its purpose. Olefirenko says that adding detail to prompts improved some results but did not provide the reliability or control she wanted for production.
#1 Best Overall
Her alternative was to generate MJML components rather than one large HTML document. MJML supplies layout abstractions, while the system defines email-specific units such as a hero, content section, image, call to action, columns and footer. Each component has a stated purpose and limits on what it can contain. Components can be generated separately, validated as smaller units and later edited in a visual builder.
This shifts the model’s task from inventing an entire email structure to selecting and filling a more explicit vocabulary. It also gives the surrounding software smaller pieces to inspect and change. The interview does not claim that component constraints eliminate rendering differences; those require separate checks.
How the quality checks fit together
The quality system developed from a practical question: is a generated email good enough? Comparing output against the rules used to generate it revealed a weakness in relying on rules alone. An email could obey known constraints and still look poor or miss its marketing purpose. The team reviewed public templates and real marketing emails as references, without assuming every existing example was good.
Recommended Free Tools
Deterministic checks catch known failures
Rules-based validation checks properties that can be expressed explicitly. Olefirenko describes checks for MJML validity, nesting, columns, attributes, section order, duplicate content, contrast, expected content blocks and requirements for particular email types. These checks make repeatable structural and content constraints testable rather than leaving them to a model’s judgment.
Semantic review asks whether the message makes sense
A language model evaluates whether the email matches its intended type, business context and campaign purpose. The interview says the assessment also draws on marketing concepts including AIDA, PAS and Cialdini’s persuasion principles. These judgments complement deterministic rules: they address fit and meaning, not just whether the markup satisfies a schema.
Visual review checks the rendered result
The rendered email is reviewed for spacing, empty areas, visual balance and available brand constraints. This catches problems that may not be apparent from the component data or semantic assessment alone. The system rolls its evaluation into an internal Email Quality Score, but Olefirenko stresses that the underlying assessment is more granular: “The main lesson for me was that no single metric tells you an email is good.” The score is an internal measure, not an independently validated benchmark.
Why a browser preview cannot prove client compatibility
Dark mode proved especially difficult because it crossed the data model, editor and generation pipeline. MJML provided layout abstractions but not a component-level light-and-dark theme model, while Apple Mail, Gmail and Outlook handle automatic color changes differently.
As Olefirenko describes it, relevant blocks store paired light- and dark-mode values in the system’s MJML representation. The generation pipeline emits a consolidated dark-mode style block before compilation. Deterministic classes, color-scheme metadata and CSS precedence help manage client-side inversion. These are engineering measures for handling variation, not a guarantee that every client will display every message identically.
The editor runs inside an iframe and can inherit the operating system’s theme. The team separated editor preview behavior from production behavior and added a manual dark-mode toggle. An initial AI-based annotation step added roughly 30 to 60 seconds to generation, according to Olefirenko; it was moved to a separate asynchronous procedure with concurrency protection. That is her report about the initial implementation, not a general estimate for dark-mode processing.
The rendering service uses Chromium. Olefirenko calls it a useful visual proxy, but cautions that it is not Outlook: “Our rendering service uses Chromium, which is useful as a visual proxy but isn’t Outlook, so we still do real test sends.” Her account also notes that sanitization can remove media queries or Outlook-specific markup, changing the HTML that reaches a client. A meaningful test send should therefore use the same HTML planned for production; a browser screenshot by itself cannot establish how the production send will render.
How business data can shape campaign recommendations
The system is extending from writing individual emails toward recommending which campaign a business should run and when. Olefirenko describes gathering context from tracking, analytics events, CRM exports and integrations. She notes that requiring customers with less technical experience to set up a proprietary tracking layer would add friction. In the interview, Square is one of the first operational integrations, and she says its flow worked end to end at that time.
A salon rebooking example
Consider a salon customer who tends to return for a service about every six months. At five and a half months, a system with the relevant history could identify the timing and audience, explain the reason for contacting that customer, and prepare a reminder campaign. This illustrates the recommendation logic; it is not a measured campaign result.
Normalizing data across business systems
An “X-ray” scan of available business and customer data establishes a baseline and looks for opportunities. Underneath it, a graph-based registry normalizes entities such as customers, services, events and campaigns across providers’ different schemas. New events after a send can enter the same representation, giving the system a common structure for reasoning about what happened.
The outcome loop was still incomplete
At the time of the interview, automatic attribution between later business events and the original campaign rationale was still being built. Opens and clicks can provide context, but a rebooking campaign ultimately aims to produce another booking. Olefirenko says scheduled reevaluation of results, with a weekly default cadence, was also in development. Her priority is the intended business outcome: “The metric I care about most is the business goal.”
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the architecture says about engineering leadership
Olefirenko says she keeps decisions that define system boundaries, affect multiple areas, are costly to reverse, materially change infrastructure costs, create partner dependencies or require broad team alignment. When goals and constraints are clear, she is comfortable delegating implementation decisions that are local and reversible.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsShe connects this approach to earlier work at airSlate, where she worked on a common schema framework for automation integrations called Add-ons and later Bots. That experience shaped a habit of identifying the entities in a problem, their attributes and relationships, and which parts are common versus case-specific. In her account, the same way of thinking informed both the email-generation vocabulary and the later abstractions for businesses, customers, services, events and campaigns.
What is implemented, and what remains a boundary
- Described as working: constrained MJML component generation, a layered evaluation approach, dark-mode data and pipeline handling, visual rendering, real test sends, and the Square integration flow.
- Described as in development: automatic attribution linking later business events to the campaign rationale and scheduled reevaluation of campaign results.
- Not established by the interview: independent validation of the internal quality score, audited campaign uplift, or a guarantee of consistent rendering across email clients.
Olefirenko also reports that selecting models for different pipeline stages, together with ontology work, reduced generation cost by roughly three times. The interview gives no baseline, calculation method or measurement period, so this should be understood as her reported result rather than an independently verified cost benchmark.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
