How avatarin built a 24/7 retail agent with GPT-Realtime

avatarin built a retail agent on OpenAI’s GPT-Realtime and put it in front of Yamada Denki shoppers for 24/7 multilingual support. In two weeks, 30,000 people used the agent, and 92% of survey responses were positive.

The product mechanics are voice-first and always on: GPT-Realtime drives the conversation so shoppers can get help outside staffed hours and across languages without a human handoff. That is the architecture signal in the deployment — realtime speech and response looped into a store-facing agent rather than a text chatbot or scheduled staff queue.

For engineers and builders, the numbers matter more than the pitch. Thirty thousand users in two weeks is enough traffic to surface latency, language routing, and fallback failure modes under real retail load. A 92% positive survey rate is a usable quality bar if you are shipping similar always-available agents and need a concrete success metric instead of anecdotal demos.

In market terms, this is a live-store deployment at Yamada Denki, not a lab pilot. avatarin is selling the agent as store infrastructure; OpenAI is the model layer underneath. The competitive angle is operational coverage — multilingual, round-the-clock support in physical retail — where human staffing does not scale cleanly.

Watch two things next: whether the same GPT-Realtime stack holds quality as traffic grows past the first 30,000 users, and whether the 92% positive score stays stable when the agent handles harder product questions and edge languages. If both hold, the pattern is a template for other big-box retailers that need 24/7 floor support without expanding headcount.