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

Today, I worked on user experience improvements on my agentic web platform by adding Tavus persona selection and fixing some vital functionalities around calls. I shipped a feature that enables portal owners to select a Tavus catalog persona or a trained clone, as well as made two important bugs fixes around calls. This task took about three hours. Although I faced some obstacles during this time, I am proud of my achievements.

The feature I added is a positive step to making the portal experience more personalized. With Tavus persona selection, portal owners can now select a persona from the Tavus catalog or a trained clone for their specific needs. This was done by adding a new dropdown field Owner edit UI for selection. Furthermore, I updated the API to include a new endpoint that lists Tavus personas. This enhances the user experience I envisioned for the agentic web, where users can engage more freely with AI.

Regardless, there were bumps on the road to this feature. I encountered some bugs and there were some serious portal call problems. I found that non-owner participants in portal calls could disrupt the meeting by making other participants perform destructive actions, like ending the meeting. To fix this, I remade the guest lockdown settings in Jitsi again so that moderators couldn’t be promoted at random, which caused confusion. I also blocked the hang up menu for guests, so they wouldn’t leave the call on purpose. I expected to be done with this sooner, but it ended up being a lot of work to test thoroughly so these changes wouldn’t be disruptive.

The other major bug I fixed was the ‘Leave Call’ button for the portal. This was an annoyance for anyone trying to exit calls. I made an update that fixed the button so that guests could leave the call without any issues. It was important to make this change to ease frustrations for an experience that centers on easy communication for users.

Creating this platform myself with AI technologies has its rewards and its challenges. I used Claude Code in my VS Code to circumvent the need for more expensive resources like Lovable or Cursor. This method saves me money and also compels me to grasp the layout and the workings of the AI on a more fundamental level. Each problem I tackle builds my expertise and brings me closer to my target of developing a one-man company with a billion-dollar valuation.

Going forward, I will need to keep improving the user experience as well as fixing bugs that come up. Building the agentic web platform is an ongoing process and I will continue iterating based on user feedback. I want to add more features that will enhance the portal, especially on the agentic SEO and GEO sides to keep the platform competitive.

To summarize, today was a day full of necessary fixes and meaningful improvements. The addition of the Tavus persona selection feature is an example of such improvement and the fixes I made have increased the platform's reliability. Tomorrow, I hope to continue facing the same challenges and improving the quality of the agentic web experience for our users.