WellnessAI

Challenges

WellnessAI aims to deliver personalized wellness coaching videos and avatars to users. However, several challenges stand in the way of achieving this goal:

– Maintaining emotional and contextual consistency in avatar delivery: The system must accurately synchronize the avatar’s visual and emotional expression with the custom-generated coaching text. Since the user writes both the avatar’s description and the program’s content, the lack of fine-grained control over the avatar’s emotional tone or lip-sync accuracy can lead to a disconnect between the coaching message and the avatar’s empathetic delivery, breaking user trust.

– Managing high inference costs and resource consumption across dual prompting: Generating both a custom visual asset (avatar) and a custom program (text) before rendering the final video involves multiple, consecutive, computationally expensive service calls. This dual reliance on paid, high-resource services, especially when combined with video processing, creates high variable operational costs and limits the scalability of personalized content production.

– Managing complex, multi-step asynchronous workflow latency for custom assets: The video creation pipeline (avatar prompt → visual asset → program prompt → text → voice→ avatar animation → final video merge) involves chaining many high-latency services across different providers. Coordinating these sequential, asynchronous steps while ensuring low latency for content delivery is a significant engineering challenge.