Three former Google X scientists aim to effectively provide a second brain virtually through an AI-powered app that gains context by listening to everything you say in the background, not sci-fi or your head meaning. Their startup, Twinmind, has raised $5.7 million in seed funding and released an Android version and a new AI speech model. There is also an iPhone version.
Co-founded in March 2024 by Daniel George (CEO) and his former Google X colleagues Sunny Tan and Mahi Karim (both CTOs), Twinmind runs into the background and captures the surrounding speeches (with user permission) to build personal knowledge graphs.
By turning spoken language, meetings, lectures and conversations into structured memory, the app can generate AI-powered notes, to-dos, and answers. It works offline, processing audio in real time to transfer the on-device and continuously capture audio for 16 to 17 hours without draining the device’s battery, the founder says. The app can back up user data, allowing you to recover conversations in the event of a device being lost, but users can opt out of it. It also supports real-time translation in over 100 languages.
Twinmind distinguishes it from AI conference note takers such as Otter, Granola and Firplies by passively capturing audio in the background all day long. To make this possible, the team built a low-level service in pure Swift, which runs natively on the iPhone. In contrast, many competitors use React natives and rely on cloud-based processing. This limits Apple’s long-term running in the background, George said in an exclusive interview.
“I spent about six to seven months just trying to continually complete this audio capture last year and find lots of hacks around Apple’s walled garden,” he told TechCrunch.
George left Google X in 2020 and won the idea for Twinmind in 2023. He served as Vice President at JPMorgan, applied for AI leads and attended consecutive meetings every day. To save time, he created a script that captured the audio, transcribed it to an iPad, and fed it to ChatGpt. Impressed by the outcome, he shared it with a friend and posted it blindly about it. That way, he built an app that could run on a personal phone, and gathered useful contexts quietly listening during meetings.
In addition to mobile apps, Twinmind offers a Chrome extension that collects additional context through browser activity. Vision AI allows you to visually scan open tabs and interpret content from a variety of platforms, including email, Slack, concepts and more.
TechCrunch Events
San Francisco
|
October 27th-29th, 2025
The startup used the extension itself as a finalist intern from over 850 applications it received this summer.
“We opened all LinkedIn profiles and CVs of 854 applicants in the browser tab and asked them to rank the best candidates in the Chrome extension,” George said. “It did an amazing job, and that’s how we hired our final four interns.”
He noted that current AI chatbots (such as Openai’s ChatGpt and Anthropic’s Claude) are unable to gather hundreds of documents and parsing into contextual information from tools such as LinkedIn and Gmail. Similarly, AI-powered browsers like Perplexity and Browser Company do not have the ability to build knowledge from offline conversations and in-person meetings.
Currently, the startup has over 30,000 users, of which approximately 15,000 are active each month. 20% to 30% of Twinmind users also use Chrome extensions, George said.
The US has been the largest Twin Minds location to date, but startups also have traction from India, Brazil, the Philippines, Ethiopia, Kenya and Europe.
Twinmind targets the general audience, but 50% to 60% of users are currently experts, about 25% are students, and the remaining 20% to 25% are used for personal purposes.
George told TechCrunch that his father is among the individuals who write autobiography using Twinmind.
One of the key drawbacks of AI is its potential to undermine user privacy. However, George argued that Twinmind is designed to work without training models on user data and without sending recordings to the cloud. Unlike many other AI note-taking apps, Twinmind cannot access audio recordings later by users – the audio is deleted on the spot – only the transcribed text is saved locally in the app, he pointed out.
Google X Experience helped speed things up
The co-founder of Twinmind worked on a variety of projects on Google X for several years. George told TechCrunch that he is working on six projects, including IYO. That experience helped the Twinmind team quickly move from concept to product.
“Google X was actually the perfect place to start your own company,” George said. “There are projects like 30-40 startups at any time. No one else, six early stage startups over two or three years, at least not in such a short period.”
Before joining Google, George worked on applying deep learning to the astrophysics of gravitational waves as part of the Nobel Prize-winning Ligo group at the University of Illinois Supercomputing Applications Center. He earned his PhD in AI at the age of 24 in just one year. This led us to the feat of participating in Stephen Wolfram’s lab in 2017 as a deep learning and AI researcher.
His early connections with Wolfram became a complete circle a few years later. He wrote his first check on Twinmind and marked his first startup investment. The recent seed round was led by streamlined ventures with participation from other investors, including Sequoia Capital and Wolfram. The round is twin values at $60 million post money.
Twin Mind EAR-3 Model
In addition to app and browser extensions, Twinmind also introduces the Twinmind EAR-3 model, the successor to the existing EAR-2, which supports over 140 languages worldwide and has a 5.26% word error rate. The new model can also recognize different speakers in conversation, with a speaker dialization error rate of 3.8%.
The new AI model is a tweaked blend of several open source models trained with curated human-released internet data, including podcasts, videos, and films.
“The more languages we support, the more the models train a wider range of speakers, so we found that the models are better suited to understand accents and local dialects,” George said.
The model will cost around $0.23 and will be available to developers and businesses through the API over the next few weeks.
Unlike ear-2, ear-3 is larger in size and runs in the cloud, so it doesn’t support a complete offline experience. However, if the internet is gone, the app will automatically switch to EAR-2 and then return to ear-3 when it comes back, George said.
With the introduction of The Ear-3, Twinmind offers a Pro subscription of $15 per month, with a context window of up to 2 million tokens and email support within 24 hours. Nevertheless, there is still a free version with all the existing features, including unlimited time transcription and voice recognition on the device.
The startup currently has a team of 11 members. They will hire several designers to enhance the user experience and set up a business development team to sell the API. Plus, there are plans to spend some money on attracting new users.
