Calendar apps are essential for productivity but it’s difficult to differentiate them enough to achieve sustainable growth from just basic use. Y Combinator supported Great poweran AI-powered note-taking software for your meetings that doesn’t involve recording bots, has reached this milestone and is now pivoting to becoming… Fabian application programming interface (API) provider so anyone can easily build an AI-powered assistant based on natural voice.
Superpowered was founded in 2020 by Jordan Dearsley and Nikhil Gupta. But after three years of working on it, Dearsley said the team wanted to work on a more challenging product. The company hasn’t discontinued the initial product as the startup said Superpowered is profitable — it’s in the process of bringing in someone to run it. Y Combinator said in June that more than 10,000 people were using the product weekly, but the company did not provide any updated numbers.
Image credits: Fabi
To date, Superpowered/Vapi has raised $2.1 in seed money from investors including Kleiner Perkins and Abstract Ventures.
Hub to Fabi
The company offers Vapi as an API to let developers create a bot using only prompts, which it then places behind a phone number. Additionally, it provides SDK integration so developers can embed the bot on websites and mobile apps.
Dearsley told TechCrunch via email that the idea to create Vapi stemmed from a personal problem. He moved to San Francisco and began missing his friends and family, who were in a different time zone. He built an artificial intelligence bot that is connected to a phone number on the other end to talk to a person in order to sort through his thoughts.
“I loved it, but I was constantly frustrated by how unnatural it was. It wasn’t like talking to someone. The voice was low, he took a long time to answer, and he interrupted me while I was talking,” he said.
“So I kept working on it and going for walks with him. Eventually, we became fascinated by this conversation problem. It’s really hard to make something sound human.” Voice assistants today It’s old-fashioned and turn-based, and we want to build something that feels human.
Technically, Vapi currently links a bunch of third-party APIs to create a powerful voice chat platform. For example, it uses solutions from Twilio for telephony, Deepgram for transcription, Daily for audio streaming, OpenAI for responses, and PlayHT To convert text to speech.
ScaleConvo, a startup in YC’s winter class of 2024, is already using Vapi to launch chatbots for sales teams and property management companies. However, Vapi has not revealed its other clients. The company is opening its application programming interface (API) with Vapi Phone and Vapi Web products today.
Challenges for Fabi
One of the startup’s biggest challenges is reducing latency, according to Magnus Revan, a former Gartner analyst and chief product officer at multimedia conversation startup Openstream.ai.
“OpenAI models need between 2 and 10 seconds to generate an answer — while the gold standard on a phone is for there to be 700 milliseconds between the user finishing speaking and then the bot starting speaking,” Revan said. “The latency is less than “One second with capable models (open source models with a large number of parameters such as the LLaMA2 70B) is very difficult.”
Currently, the Vapi has a latency of 1.2 to 2 seconds depending on various factors. Dearsley expects latency to be reduced to less than one second in the next month thanks to Vapi’s work and OpenAI improvements.
Mohamed Misbah, an angel investor in Vapi, said the startup’s solution will improve with the overall progress of the application programming interface (API).
“As OpenAI and others refine their models, the Vapi platform will become more powerful, equipped with better knowledge bases, code execution capabilities, and greater context windows. Vapi’s focus on solving the biggest areas of friction in voice communications will be an advantage as user demand for voice assistants grows.”
However, this puts the onus on improving other solutions rather than the Vapi itself. The reliance on other APIs makes Vapi less defensible if larger companies start moving into that area, Dearsley said. However, the team said it has an advantage in terms of building infrastructure to handle thousands of calls simultaneously. Dearsley confirmed that as Vapi’s web and phone APIs go public, the team will also look to build its own prototypes for voice-to-voice solutions.