French startup Rounded believes that AI voice agents will become the default way for customers to interact with businesses, and rather than building an out-of-the-box AI voice agent, it has created an orchestration platform that allows businesses to build their own voice agents. We are building a ration platform.
Rounded began developing its Web3 product and then shifted its focus to exploring AI voice agents in June 2023. “The idea was that after transcription, it would be worthwhile to just put ChatGPT in front of the synthesizer,” says the co-founder. Aymeric Vaudelin (pictured above, first from left) told TechCrunch.
However, the team soon faced the familiar conundrum of product-market fit. “After a few months, we realized that the market wasn't ready to hear about voice agents yet, so we built the product, packaged everything up and created our first agent,” Vaudelin said. I added.
The result of that effort is Donna, an AI voice agent for anesthesiologists. It seems a little random, but the startup chose that market because anesthesia secretaries have to deal with a large number of patients and are usually a highly transactional experience.
In France, when booking a surgical procedure, you must consult your anesthetist in advance to check if you have any allergies or potential complications to anesthesia products.
Anesthesia secretaries have to handle a large number of phone calls with great ease. Typically, people just want to know when an anesthesiologist is available to schedule an appointment or change a date.
Additionally, these are not sales calls, so the AI agent doesn't need to be persuasive or super competent. “In the early days, we suffered from delays of four, five, six seconds in some cases,” Vaudelin said.
Nevertheless, Rounded, with Donna's help, was able to persuade 15 private hospitals to use AI voice agents to answer their calls, and the company says the agents have been used in dozens of cases so far. It is said to have processed 10,000 conversations. Over time, Rounded has improved the product, making it more accurate, better integrated with other products, and, importantly, faster.
“For example, web call latency is now less than 700 milliseconds and closer to 600 milliseconds. Telephone connections add more or less 200 milliseconds,” Vaudelin says.
More recently, Rounded expanded its remit with an orchestration product that other companies can use to build their own voice agents.
Rounded offers a selection of ready-made AI models, including speech synthesis models, LLMs, and text-to-speech models. For example, for your first voice agent, you can use Azure for call transcription, GPT-4o mini as your LLM, and Celebrities as your speech-to-text engine.
The platform then helps define the instruction trees and prompts that make LLM work for specific use cases. “Implementing an agent means finding the right prompts, the right parameters, and the right variables within the prompts,” Vaudelin says.
“Our pitch is that we enable anyone to create great prompts and great agents. And our product is an iterative process to create highly reliable agents. We support them,” Vaudelin said.
Rounded has raised €600,000 (approximately $620,000) from UC Berkeley deep tech accelerator SkyDeck and several business angels. However, given that artificial intelligence is still a very active industry, we are confident that the startup will raise even more funding in the coming months.