Zapia solving "real world tasks" for millions with Google Cloud and Gemini

Video Description

Title: Zapia's executive assistant: Solving "Real World Tasks" for millions with Google Cloud and Gemini Featuring: Nicolas Loeff, Founder and CTO at Zapia AI. Executive Summary: Nicolas Loeff, Founder and CTO of Zapia AI, discusses how Zapia uses Gemini models and Google Cloud Platform (GCP) to provide a concierge agent that works over messaging platforms. Zapia's agent is designed to help users solve problems that cannot be addressed by searching the web and instead require talking to other people. Zapia is built entirely on GCP, utilizing services like Google Cloud Run, and is currently scaled to millions of users. Challenge: Zapia focuses on solving "real-world tasks" for its users—problems that are too complex to be solved with a simple web search and require human interaction to resolve. The company needed a technology stack that could support an agent handling these complex issues and be able to scale from ten users to millions without redesigning the system. Solution: Zapia built its entire technology stack on top of GCP, using Cloud Run for running services and designing the system for massive scalability from the start. They utilize the full spectrum of Gemini models, from Flash to Pro, by using the right model for the right problem. For example, they use Pro models for situations that involve more reasoning, while moving to the light models for simple user chatting. Nicolas notes that the Gemini models offer the best performance-to-price ratio of any model available. Result: With the help of Google Cloud, Zapia is successfully solving real-world problems for billions of users in Latin America. Success for Zapia is defined by users spending less time solving their issues while simultaneously receiving a better, cheaper service than they could find elsewhere. User satisfaction is high, with over 92% of conversations rated positively. The service's quality is also reflected in the app's high rating, which is 4.8. Interview highlights and key takeaways from our interview with Nicolas Loeff, CTO at Zapia AI: Zapia's success and ability to scale are deeply reliant on the architecture of Google Cloud Platform (GCP). By strategically using the full spectrum of Gemini models, they optimize for tasks requiring more reasoning (Pro models) versus simple user chatting (light modes), citing that Gemini offers the "best performance to price ratio of any model out there". Success for Zapia means users "are spending less time solving these issues and at the same time they're getting a better service that is cheaper than what they could find". The company consistently sees over 92% of conversations rated positively and has an app rating of 4.8. Zapia is currently "solving real world problems for our millions of users in Latin America". Google Cloud products used: Gemini models, Google Cloud Run