Free tools Windows power users keep installed
One-click scans. No signup required.
Reliable AI-generated email takes more than a detailed prompt. Alina Olefirenko, co-founder and CTO of AlpacaRelay, describes replacing free-form HTML output with constrained MJML components, then checking generated campaigns with structural rules, semantic review, rendered previews, and real test sends. Her account also draws an important line: connecting campaign recommendations to business outcomes was still a work in progress when she spoke with The AI Journal on 5 October 2026.
Why prompt-only email generation falls short
Email is an unusually unforgiving format for generated content. Clients vary in their support for HTML and CSS, and many email layouts still depend heavily on tables. A model can produce markup that looks plausible in one preview yet behave unpredictably in a recipient’s inbox.
As an Amazon Associate I earn from qualifying purchases.
Olefirenko says more detailed prompts improved some results, but prompt-only HTML generation did not provide the reliability and control she wanted for production. As she put it, “A model can produce HTML that looks reasonable without producing an email I would actually trust in production.”
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 →Constrain generation with reusable components
Instead of asking a model to create one large HTML document, Olefirenko’s team designed generation around MJML components. The vocabulary includes elements such as a hero, content sections, images, calls to action, columns, and a footer.
#1 Best Overall
Each component has a defined purpose and limits on what it can contain. Components can be generated separately, validated as smaller units, and edited afterward in a visual builder. This gives the model clearer boundaries and gives the product more control over what reaches the final email.
Olefirenko connects this approach to her earlier work at airSlate, where she helped build a common schema framework for automation integrations called Add-ons and later Bots. She says that experience shaped her habit of defining entities, attributes, relationships, and the distinction between shared and case-specific data. The same way of thinking informed both the email-generation vocabulary and later abstractions for businesses, customers, services, events, and campaigns.
Use several kinds of quality checks
The team’s evaluation grew from a practical question: is a generated email good enough? Comparing emails with the rules used to create them revealed a weakness in rule-only checking. An email can satisfy known constraints and still look poor or miss its purpose. The team reviewed public templates and real marketing emails as references, without treating every example as a model to copy.
Deterministic checks catch structural problems
Rules can verify whether MJML is valid, components are nested properly, columns and attributes are used correctly, and sections appear in an expected order. Other checks can flag duplicate content, contrast issues, missing content blocks, or violations of requirements for a particular email type.
Semantic review checks intent and context
An LLM can assess whether an email fits its intended type, business context, and campaign purpose. The interview also describes bringing concepts such as AIDA, PAS, and Cialdini’s persuasion principles into assessment of marketing effectiveness. These judgments address meaning and fit; they do not replace structural validation.
Visual review checks the rendered result
Rendered previews help identify spacing problems, empty areas, poor visual balance, and mismatches with available brand constraints. This catches issues that are hard to see from the component structure alone.
The layers roll up into an internal Email Quality Score, but Olefirenko says 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 a product-internal aid, not an independently validated benchmark.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Dark mode requires more than a preview switch
Dark mode was the most difficult rendering case described in the interview because it affected the data model, editor, and generation pipeline at once. MJML provides layout abstractions, but not a component-level light-and-dark theme model. Apple Mail, Gmail, and Outlook also handle automatic color changes differently.
The described implementation stores paired light- and dark-mode values on relevant blocks, retains them in the MJML representation, and emits a consolidated dark-mode style block before compilation. Deterministic classes, color-scheme metadata, and CSS precedence help manage client-side color inversion.
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 in the team’s reported experience; Olefirenko says it was moved to a separate asynchronous procedure with concurrency protection. That timing describes the initial implementation, not a general dark-mode processing benchmark.
Browser previews are useful, but test sends matter
The rendering service uses Chromium. Olefirenko calls it a useful visual proxy, while emphasizing 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.”
A screenshot can reveal visual defects in the browser-based rendering path, but it cannot establish how every email client will handle the message. Sanitization may change the HTML that reaches a client, including by removing media queries or Outlook-specific markup. For that reason, the interview says test sends use the same HTML planned for production. The production path—not just the editor preview—needs checking.
Move from composing emails toward campaign recommendations
Olefirenko describes an effort to recommend which campaign a business should run and when, rather than only composing individual messages. The system can draw context from tracking, analytics events, CRM exports, and integrations. She notes that building a proprietary tracking layer would create extra setup for less technical customers.
Square was one of the first operational integrations she discussed, and she said that flow worked end to end at the time of the interview. The example she gives is a salon customer who tends to return for a service about every six months. If five and a half months have passed, the system could identify a likely timing, audience, reason for contact, and a ready-to-send reminder. This illustrates the proposed logic; it is not evidence of a measured campaign result.
An “X-ray” scan of available business and customer data builds a baseline and looks for opportunities. Beneath that, a graph-based registry normalizes entities—including customers, services, events, and campaigns—across provider schemas. New events after a send enter the same representation.
Business outcomes are the intended feedback signal
Opens and clicks can add context, but they do not necessarily answer whether a campaign accomplished its purpose. For a rebooking reminder, the business outcome is whether the customer booked again. “The metric I care about most is the business goal,” Olefirenko says.
At the time of the interview, the connection between a campaign’s original rationale and later business events was still being built. Scheduled reevaluation of results, described with a weekly default cadence, was also in development. The account therefore describes a direction for the feedback loop, not a completed, measured attribution system.
Which engineering decisions should stay centralized?
Olefirenko says she retains decisions that define system boundaries, affect multiple areas, are costly to reverse, materially change infrastructure costs, create partner dependencies, or require broad team alignment. Once the goals and constraints are clear, she is comfortable delegating implementation choices that are local and reversible.
She also reported a roughly threefold reduction in generation cost, attributing it to selecting models for different pipeline stages together with ontology work. The interview provides no baseline, calculation method, or measurement period, so the figure is her reported result rather than an independently verified cost comparison.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallThe engineering pattern in her account is to constrain what the model can generate, validate different failure modes with different methods, and check the actual production send path. The system’s next challenge is to connect campaign intent reliably to later business events and use those outcomes to improve future recommendations.
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.
