GIGI Voice Fixes & Map Search Enhancements — May 16, 2026

I spent about 21 hours in one of those long stretches in the life of a solo developer. I have been working on a number of bugs and changes focused on getting an improved user experience and performance with GIGI's voice functionality and the map search experience. Today, I published fixes for GIGI's voice pitch issues, improvements to the map search delays, and minor improvements to the sponsored keywords functionality.

I had to handle the voice glitch first because a lot of users complained about it. It was really weird since there was nothing else that was fast or high pitched like a chipmunk for example and then the greeting GIGI gave at the start was like that. I had to self-host the libsamplerate worklet to figure that out. It took me a long time to figure out what the glitch was, but I finally got it fixed so GIGI should sound a lot more natural now.

The next thing I focused on was improving the user experience in the map search function. There had been a lag of almost 3 seconds and a 10 second delay in updating the URLs after a user performs a search. With super fast URLs this is a lag unacceptable, as I optimized several things. The first of which was ensuring that duplicate fetch calls are made at the map resolve effect (the Nav bar had prior populated the keyword stakes). The second one was modifying the fly-to animation and how the url updates were handled so that during zoom actions the URL does not flash too much ( as to update the URL during the fly to action). I think the map search feels and functions better than before after these optimizations.

While performing these optimizations when improving the map search, I took the opportunity to also resolve other unrelated issues like the bugs that cause the URL to flicker and the issues surrounding the silent search. I improved the experience of zooming since it is now less jerky, and the flickering during zoom actions is much less due to the new structure of sequential URL updates that occur during zoom actions to a single routed replaced call. I also updated the system to inform users when their search returned no results so that they don't feel that their search went unnoticed. To do this, I added a toast message at the center top of the search results.

Updates have been made on the sponsored keywords feature. I resolved an issue that resulted in the Chrome extension not displaying fiat rows correctly due to duplicate prefixes. Little things really do affect the credibility of the platform and impact the user experience positively. I also set up GIGI to autofill lead forms as well as upsell opportunities to make new user onboarding easier.

Over the course of the 21-hour marathon, I learned that the task of diagnosing the GIGI voice glitch took longer than I had planned because I had to locate the exact library that was causing the problem. The same goes for the map search optimization; I was not able to achieve the expected results in the early optimization attempts. Despite these obstacles, I learned the importance of comprehensive user testing as well as the value of accessing the user experience from multiple points.

During this day, I took a moment to consider the weight of ambitious long-term goals of building this project for me and AI. Although I could have used a paid tool like Lovable or Cursor, I chose to use Claude Code in VS Code to not spend money and for me to learn more about the codebase. I hope to further my goal of building a one-man show with a billion-dollar goal with each fix and improvement.

When reflecting on today’s work, I feel pride for the progress, while also appreciating the work still to come. Each feature shipped and bug squashed is another milestone in the journey. Each commit is not just another step in the creation of the platform. It is also the articulation of the vision for what the agentic web can be. Tomorrow will bring new challenges, and I look forward to meeting them.