The purpose of three former Google X scientists is to give you a second brain in practice-Science Fi or your head does not mean. Their starting, TwinThe seeds have collected 7.7 million in financing and have released the Android version, with a new AI speech model. It also has an iPhone version.
In March 2024, Daniel George (CEO) and his ex -Google Xi colleagues were jointly founded by Sunny Tang and Fisheries Cream (both CTOs), the Twin Mind runs in the background, and occupies a paid speech (with user permission) to create a personal graph.
Speaking ideas, meetings, lectures and conversations into structural memory, the app can produce AI -powered notes, todas and answers. The founders say, it works offline, follows the audio so that they can process audio in real time for audio copy, and can take the audio for 16-17 hours without extractioning the device’s battery. The app can also back up the user’s data, so if the device is lost, the conversation can be recovered, though the user can opt out. It also supports real -time translation into more than 100 languages.
Twin Mind meets himself with AI meeting notes, such as Outra, Greenola, and Fireflies all day long, catching audio in the background. To make it possible, the team served a lower level in pure Swift that operates locally on the iPhone. On the contrary, many competitors use local reactions and rely on cloud -based processing, which prevents Apple from running in the background for expanded periods, George said in an exclusive interview.
He told the Tech Crunch, “We spent about Six six to seven months last year to complete the capture of this audio only and arrived there to find a lot of hex around the Apple wall garden.”
George left Google X in 2020 and came to the idea for Twin Mind in 2023 when JP was working as vice president in Morgan and applied AI Lead, and attended back -to -back meetings every day. For the time -savings, he created a script that occupied the audio, copied it to his member, and fed it in a chatigue – who started to understand his plans and even developed a usable code. Inspired by the results, he shared it with friends and posted on the blind, where others expressed interest but did not want to run anything on their laptop. Because of this, she made an app that can run on a personal phone, quietly listening to meetings to collect useful contexts.
In addition Mobile appOffers Twin Mind A Chrome extension Which collects additional context through browser activity. Using Vision AI, it can scan open tabs and translate content from different platforms, including email, slack and idea.
Taxkarnch event
San Francisco
|
October 27-29, 2025
Even this summer, this extension was used to shortlisted interns from more than 850 applications.
“We opened 854 applicants’ LinkedIn profiles and CVs in the browser tabs, then asked Chrome Extension to classify the best candidates,” George said. “He did an amazing thing – so we hired our last four interns.”
He noted that hundreds of documents or pars sign-ups from the current AI chat boats, including Openai’s chatgot and anthropic cloud-saucet information information, could not easily be processed. Similarly, people from AI -powered browsers, such as troubles and browser company, are capable of producing your offline conversation and personal meetings.
Startups currently have more than 30,000 users, of which about 15,000 are active every month. George said 20-30 % of the twin-minded users also use Chrome Extension.
Although the United States is still the largest base in the Twin Mind, startup is also detected from India, Brazil, the Philippines, Ethiopia, Kenya and Europe.
Twin Mind targets ordinary audiences, although 50-60 % of its users are currently professional, about 25 % of students, and the remaining 20-25 % are individuals who are using it for personal purposes.
George told Tech Crunch that his father was among the people who use the Twin Mind to write his autobiography.
One of the major flaws of AI is the ability to compromise the user’s privacy. But George emphasized that the Twin Mind does not train its models on user data and is designed to work without sending a recording on the cloud. He noted that unlike many other AI note-taking apps, Twin Mind does not allow users to later access audio recording-the audio is deleted on the fly-while the only copied text is saved locally in the app.
Google X experience helped to speed up things
Twin Mind’s co -founders spent a few years working on various Google X projects. George told Tech Crunch that he only worked on six projects, including iyo -Ai -powered Airbuds team, who recently made headlines to prosecute Open and Jonny Io. This experiment helped the Twin Mind team imagine moving the product quickly.
George said, “Google X was actually the best place to start your own company.” At any time, projects like 30 to 40 are taking place. No one else is found to work at the beginning of the sixth stage in two or three years before starting to their own-at least not in the short term. “
Prior to joining Google, George worked to apply deep -leaving to the astronomical physics of the gravity wave as part of the winning Lego group at the University of Illinois National Center for Supercamping Applications. He had completed a PhD in AI for Astro Physics – at the age of 24 – in just one year – a feat that led to a deep learning and an AI researcher in 2017 in Stephen Wolfram’s research lab.
The initial relationship with Wolfram came into full circle after years-he finished writing the first check for the Twin Mind, which identified his first investment in the startup. The recent seed round was led by smooth ventures, with the participation of other investors, including the capital of Satonia and Wolfram. The round is priced after $ 60 million.
Twin Mind Ear-3 Model
In addition to the expansion of its apps and browsers, the Twin Mind has also introduced the Twin Mind Air -3 model, which is the successor of its current Air2, which supports more than 140 languages worldwide and its literal error rate is 5.26 %. The new model can also recognize different speakers in the conversation and it has a speaker directing error rate 3.8 %.
The new AI model is a good combination of several open source models, trained on a set of human unannounced Internet data-which includes podcasts, videos and films.
George said, “We have learned that the more languages you cooperate, the better the model is better than understanding the tone and regional dialects because it is a broader range of speakers.”
The model price is 3 0.23/ hr and Will be available through an API In the next few weeks, developers and businesses.
Unlike EAR-3, EAR-2, does not support full offline experience, as it is larger in size and runs on the cloud. George said, however, the app automatically changes to EAR-2 if the Internet ends and then returns to EAR-3 when he returns.
With the introduction of EAR-3, the Twin Mind now offers a pro-subscription in $ 15/month, with a major context of up to 2 million token and email support within 24 hours. Nevertheless, the free version still exists with all current features, including unlimited times of duplication and on -device speech.
Startup currently has a team of 11 members. It plans to hire some designers to enhance their user experience and create a business development team to sell its API. In addition, there are plans to spend some money on the acquisition of new consumers.