GIGIlistening…

Build Journal

Tavus Persona Improvements & Data Layer Enhancements — June 5, 2026

I tackled Tavus persona improvements and data layer enhancements today, shipping features and fixes to boost agentic web performance and user experience.

15 changes3 min readby Rob

What shipped

  • Tavus Persona FixPrevented re-deriving owned personas to stabilize management.
  • Keyword ManagementAdded pause and delete options for paid keywords.
  • Dashboard OverhaulIntroduced 'Overview' and 'Keywords' tabs for clarity.
  • Data Layers ExpansionAdded 14 new unlockable data layers for richer insights.
  • Map UI ImprovementsEnhanced visuals and accessibility for map elements.

Today, I set out to enhance the Tavus persona management and improve the data layers functionality within my spatial AI platform. After a long 10-hour day, I successfully shipped 12 feature improvements and 3 critical fixes, each contributing to a more robust and user-friendly interface. These updates are not just about adding features; they are about refining the foundations of the agentic web experience I am building at hashtag.org.

One of the significant improvements was addressing the Tavus persona system. I implemented a fix to prevent the system from re-deriving an already-owned persona, which was causing issues with persona replication. This bug had been a thorn in my side, as it led to broken replicas being inherited on subsequent calls. To resolve this, I added a name-prefix guard, ensuring that once a persona is created, it is treated as terminal. This change not only stabilizes the persona management but also streamlines the user experience, making it more intuitive.

Next, I dove into the keywords management system. I introduced a pause and delete functionality for each paid keyword, which allows users to deactivate billing temporarily or remove subscriptions entirely. This feature is essential for users who want more control over their keyword investments and aligns with my goal of making the platform as user-centric as possible. I had to deal with some tricky logic to ensure that the new endpoint for managing keywords operated seamlessly, but it was worth the time spent.

The dashboard also received a significant overhaul today. I split the portfolio dashboard into two distinct tabs: 'Overview' and 'Keywords.' This change was driven by feedback indicating that users wanted a cleaner interface that made it easier to scan through critical information. By separating these sections, I believe I've made it simpler for users to navigate and find what they need quickly.

Another area I focused on was the data layers. I expanded the Apify catalogue to include 14 new unlockable layers, which enrich the data available to users significantly. Each layer provides valuable insights, such as local leads, Google reviews, and job postings from various platforms. Implementing this feature took considerable effort, especially ensuring that the system can handle bad actor slugs without breaking the scan process. The new layers are designed to auto-surface in the unlock modal and scan pipeline, making them easily accessible to users.

I also worked on enhancing the map interface. I made visual improvements, such as changing the color of in-flight and orbiting drones from fuchsia to yellow, which I believe is a more visually appealing choice. Additionally, I removed redundant tooltips that cluttered the interface, opting instead for aria-labels to improve accessibility for screen reader users. These small UI tweaks might seem minor, but they contribute significantly to the overall user experience and make the product feel polished and professional.

Throughout the day, I encountered a few bugs that ate into my time, particularly with the data layers and the Tavus persona system. Diagnosing these issues required a deep dive into the codebase, and while frustrating, it was a valuable learning experience. Each fix not only resolved immediate problems but also provided insights into improving future workflows. I find that working solo with my AI assistant, Claude Code, has allowed me to iterate quickly and efficiently, avoiding the overhead of collaborative tools that might slow me down.

As I reflect on today's progress, I feel a sense of pride in how far the platform has come. Each feature and fix I implemented today brings me one step closer to my vision of building a one-man-show company with a billion-dollar valuation. The combination of Tavus persona enhancements and expanded data layers positions the agentic web to deliver even more value to users, reinforcing my commitment to creating a superior product. With each commit, I am not just shipping features; I am laying the groundwork for a future where AI and user experience converge seamlessly on hashtag.org.

← Back to the full build journal