The Evolution of Superpowered: Introducing Vapi, Your Voice-Based AI Assistant
Calendar applications play a crucial role in enhancing productivity, yet it can be challenging for companies in this space to achieve sustained growth beyond their core functionalities. Superpowered, an AI-driven notetaking tool for meetings—backed by Y Combinator—encountered this very challenge and is now transforming into Vapi, an API provider designed to help developers seamlessly create natural-sounding, voice-based AI assistants.
Founded in 2020 by Jordan Dearsley and Nikhil Gupta, Superpowered spent three years refining its notetaking product. Dearsley revealed that the team sought a more ambitious project and, while they are not discontinuing the original product—having achieved profitability—they are actively looking to appoint a new leader to oversee it. As of June, Y Combinator noted that more than 10,000 users were engaging with Superpowered weekly, although updated figures have not been shared.
From Superpowered to Vapi
Superpowered/Vapi has successfully secured $2.1 million in seed funding from notable investors, including Kleiner Perkins and Abstract Ventures.
Vapi is tailored as an API that enables developers to effortlessly create conversational bots using simple prompts linked to a dedicated phone number. The platform also provides an SDK for seamless integration of bots on websites and mobile applications.
Dearsley shared via email that Vapi's inception was born from a personal challenge. After relocating to San Francisco, he found himself missing his friends and family in different time zones. To cope, he built an AI bot that could facilitate conversations and help him organize his thoughts.
He shared, “While I appreciated the concept, I grew increasingly frustrated by its lack of naturalness. The interactions felt robotic—the voice was off, response times were lengthy, and it often interrupted me.” This frustration fueled his desire to enhance the user experience.
During his walks, Dearsley further explored the complexities of human-like conversation. “Creating a genuinely human-like experience is a significant challenge. Current voice assistants often feel cumbersome; our goal is to develop a solution that mimics real human interaction.”
Powering Conversations with Technology
Vapi integrates a variety of third-party APIs to create a powerful voice interaction platform. It leverages Twilio for telephony, Deepgram for transcription, Daily for audio streaming, OpenAI for conversational responses, and PlayHT for text-to-speech functionalities.
ScaleConvo, a startup participating in Y Combinator’s winter 2024 batch, is already utilizing Vapi to develop conversational bots tailored for sales teams and property management companies. However, Vapi has not disclosed information about additional clients. The company is officially launching its API, featuring Vapi Phone and Vapi Web products.
Navigating Challenges
One of Vapi's pressing challenges, as noted by Magnus Revan, a former Gartner analyst and chief product officer at Openstream.ai, is minimizing latency. “OpenAI models typically require 2-10 seconds to generate a response, whereas, in a phone conversation, the ideal is around 700 milliseconds from when a user finishes speaking to when the bot starts. Achieving sub-1-second latency with advanced models like LLaMA2 70B is incredibly challenging,” Revan explained.
Vapi currently experiences a latency range of 1.2 to 2 seconds, but Dearsley aims to reduce this to under one second within the next month, benefiting from improvements both in Vapi’s own technology and OpenAI's advancements.
Mohamed Musbah, an angel investor in Vapi, noted that as improvements in APIs occur, Vapi’s solutions will also enhance. “With advancements from OpenAI and similar platforms, Vapi will strengthen its capabilities, providing better knowledge bases and more extensive context windows. Focusing on reducing friction in voice communication will become increasingly valuable as voice assistant demand rises,” he stated.
However, this dependency on external APIs raises concerns about Vapi's long-term defensibility, as larger companies may encroach on this space. Nonetheless, Dearsley assured that Vapi boasts a robust infrastructure capable of managing thousands of simultaneous calls. With the introduction of its web and phone API to the public, the team plans to develop proprietary models focused on audio-to-audio solutions.
Ranking
