Tavus Persona Selection & Portal Call Fixes — May 20, 2026

Today, I set out to enhance the user experience on my agentic web platform by implementing Tavus persona selection and fixing some critical call functionalities. I successfully shipped a feature that allows portal owners to choose between a Tavus catalog persona or a trained clone, alongside two important bug fixes that improve overall call management. This work took me about three hours, and while I'm proud of what I accomplished, the journey was not without its challenges.

The new feature I added is a significant step toward personalizing the portal experience. With the Tavus persona selection, portal owners can now choose a persona that best fits their needs, whether it's a pre-defined Tavus catalog persona or a trained clone tailored for their specific use case. I implemented this by adding a new dropdown in the owner edit UI that allows easy selection of the desired persona. Additionally, I enhanced the API with a new endpoint that retrieves the available Tavus personas, allowing for seamless integration into the existing system. This feature not only enriches the user experience but also aligns with my vision for an agentic web where users can interact with AI in a more personalized manner.

However, the path to this feature was not entirely smooth. I encountered a few bugs that required immediate attention. The first issue was related to the portal's call functionality. I found that non-owner participants in portal calls were able to perform actions that could disrupt the meeting, such as ending the meeting or making destructive changes. To fix this, I re-applied the Jitsi guest lockdown settings, ensuring that moderators could not be promoted randomly, which previously led to confusion. I also intercepted the hangup menu for guests, preventing them from accidentally leaving the call incorrectly. These fixes took longer than I anticipated, consuming a significant portion of my time and requiring careful testing to ensure stability.

The second bug I addressed was related to the 'Leave Call' button in the portal. It was not functioning as intended, which could lead to frustration for users trying to exit calls. I implemented a fix that restored the button's functionality, ensuring that guests can leave the call smoothly without any issues. This fix was crucial for maintaining a user-friendly experience, especially in a platform designed to facilitate seamless communication.

Reflecting on today's work, I realize that building this platform solo with AI assistance is both rewarding and challenging. I leveraged Claude Code in VS Code to tackle these issues without relying on expensive tools like Lovable or Cursor. This approach not only saves costs but also forces me to deepen my understanding of the platform's architecture and the intricacies of the AI components. Each problem I solve adds to my skill set and moves me closer to my goal of building a one-man-show company with a billion-dollar valuation.

As I look ahead, I see the need to continue refining the user experience and addressing any bugs that may arise. The journey of building the agentic web platform is an ongoing process, and I’m committed to iterating and improving based on user feedback. I also plan to explore additional features that could enhance the portal's capabilities, particularly in areas that align with agentic SEO and GEO strategies, ensuring that the platform remains competitive in the market.

In conclusion, today was a productive day marked by meaningful enhancements and necessary fixes. The addition of the Tavus persona selection feature is a notable step forward, and the fixes I implemented have strengthened the platform's reliability. I look forward to tackling more challenges tomorrow and continuing to build a robust agentic web experience for users.