Calendar apps are essential to productivity, but it’s difficult to differentiate them enough from their core use cases to sustain growth. Powered by Y Combinator super powerfulan AI-powered meeting note-taking tool that does not require bot recording, has overcome this obstacle and is currently Vapian API provider, makes it easy for anyone to create natural, voice-based, AI-powered assistants.
Superpowered was founded in 2020 by Jordan Deersley and Nikhil Gupta. But Dearsley said after three years of work, the team wanted to work on a more challenging product. Superpowered is profitable, the startup said, and the company has no intention of shutting down its first product and is in the process of hiring someone to run it. Y Combinator announced in June that more than 10,000 people use the product each week, but the company did not provide updated numbers.
Image credits: Vapi
To date, Superpowered/Vapi has raised $2.1 in seed money from investors including Kleiner Perkins and Abstract Ventures.
Pivot to Vapi
The company offers Vapi as an API that allows developers to create bots using only a prompt and putting it behind a phone number. Additionally, SDK integration is also provided, allowing developers to embed bots on their websites and mobile apps.
Dearsley told TechCrunch via email that the idea to build Vapi stemmed from a personal problem. He moved to San Francisco and began to miss his friends and family in different time zones. He built his AI bot that connects to the other party’s phone number to have a conversation with someone to organize their thoughts.
“I liked it, but I was constantly annoyed by how unnatural it was. It didn’t feel like talking to a person. The audio cut out, it took a long time to respond, and when I was talking it would interrupt me. ” he said.
“So I kept working on it and went for walks with it. Eventually, we fell in love with this conversation problem. It’s really hard to make something feel human. Voice assistants. today It’s clunky and turn-based, so I wanted to create something with a human touch. ”
Technically, Vapi is currently integrating a number of third-party APIs to build a robust voice conversation platform. For example, we use solutions from Twilio for phone calls, Deepgram for transcription, Daily for audio streaming, and OpenAI for responses. PlayHT For text reading.
ScaleConvo, a 2024 YC Winter Batch startup, is already using Vapi to launch conversational bots for sales teams and property management companies. However, Vapi did not reveal other customers.The company publishes his API Current Vapi Phone and Vapi Web products.
Vapi challenges
Magnus Revan, former Gartner analyst and chief product officer at multimodal conversation startup Openstream.ai, says one of the startup’s biggest challenges is reducing latency.
“OpenAI models take between 2 and 10 seconds to generate a response. On the phone, the gold standard is 700 milliseconds between when the user finishes speaking and when the ‘bot’ begins speaking. And with capable models (high parameter count open source models like LLaMA2 70B) it is very difficult to achieve sub-second latencies,” he says.
Currently, Vapi’s latency is between 1.2 and 2 seconds depending on various factors. Dearsley expects that Vapi’s own efforts and improvements to his OpenAI will bring latency down to below a second for him in the next month.
Mohamed Musbah, an angel investor at Vapi, also said that the startup’s solution will be improved by overall advances in APIs.
“As OpenAI and other companies improve their models, Vapi’s platform will become more powerful, with a better knowledge base, code execution capabilities, and a larger context window. “As demand grows, Vapi’s focus on solving the biggest friction areas in voice communications will be an advantage,” he said.
However, this puts the responsibility on improving other solutions, not Vapi itself. Dearsley said the reliance on other APIs will make Vapi less defensive if larger companies start moving into the space. But the team said it has an advantage in that it has built the infrastructure to handle thousands of calls simultaneously. Dearsley emphasized that with the general availability of Vapi’s web and phone APIs, the team will also look to build proprietary models for his audio-to-audio solutions.