From Prompt-and-Response to Agentic Workflows: Engineering an AI Content Platform for Telehealth

A healthcare organization approached us with an AI-powered platform designed for SEO. While the prototype functioned effectively, it struggled to scale due to its architecture. The system relied heavily on a single AI provider and lacked the capability to handle multiple brands.

The challenge was to transform the prototype into a production-ready software without compromising the workflow that had been validated. We identified several key issues: the original architecture's limitations, the dependency on a single AI model, the absence of persistent context, the lack of human evaluation, and the need for compatibility with various publishing environments.

To address these issues, we restructured the workflow into a more complex, multi-stage process. Each stage had a specific responsibility, moving away from treating the LLM as a general-purpose assistant. We shifted the AI layer to OpenRouter, enabling the system to work with multiple models through a common integration. An additional layer for image generation was also incorporated.

Persistent context was introduced to the workflow, allowing the system to remember previous stages and maintain continuity. This enabled different agents to work on various stages while retaining the necessary information. Despite the AI-generated content, human evaluation remained crucial, particularly in a domain like healthcare where quality is paramount. Human reviewers provided natural-language feedback, which became part of the system's context for future generations.

Recognizing the need for compatibility with multiple publishing environments, we added CMS capabilities to the platform. This allowed the system to support both WordPress and Astro-based websites while maintaining compatibility with WordPress. Custom WordPress components were also created for specific generated content, such as HTML-based infographics.

The final result was a multi-tenant platform supporting around 14 brands. The system connected various stages, including discovery, research, AI agents, SEO/AEO/GEO evaluation, human feedback, CMS, and publishing. The key takeaway was not to use AI agents, but to understand when a prototype has progressed beyond the experimental stage.

Once an AI workflow becomes part of a real healthcare business, reliability, context, model abstraction, human review, and scalable architecture must be considered as fundamental product requirements.