Ex-google X Trio Wants Their Ai To Be Your Second Brain — And They Just Raised $6m To Make It Happen

9 hours ago

Three erstwhile Google X scientists purpose to springiness you a 2nd encephalon virtually — not successful nan sci-fi aliases chip-in-your-head consciousness — but done an AI-powered app that gains discourse by listening to everything you opportunity successful nan background. Their startup, TwinMind, has raised $5.7 cardinal successful seed backing and released an Android version, on pinch a caller AI reside model. It besides has an iPhone version.

Co-founded successful March 2024 by Daniel George (CEO) and his erstwhile Google X colleagues Sunny Tang and Mahi Karim (both CTOs), TwinMind runs successful nan background, capturing ambient reside (with personification permission) to build a individual knowledge graph.

By turning spoken thoughts, meetings, lectures and conversations into system memory, nan app tin make AI-powered notes, to-dos, and answers. It useful offline, processes audio successful real-time to transcribe on-device, and tin seizure audio continuously for 16–17 hours without draining nan device’s battery, nan founders say. The app tin besides backmost up personification information truthful conversations tin beryllium recovered if nan instrumentality is lost, though users tin opt-out of that. It besides supports real-time translator successful complete 100 languages.

TwinMind differentiates itself from AI gathering note-takers for illustration Otter, Granola, and Fireflies by capturing audio passively successful nan inheritance each day. To make this possible, nan squad built a low-level work successful axenic Swift that runs natively connected nan iPhone. In contrast, galore competitors usage React Native and trust connected cloud-based processing, which Apple restricts from moving successful nan inheritance for extended periods, George said successful an exclusive interview.

“We spent astir six to 7 months past twelvemonth conscionable perfecting this audio seizure continuously and getting location to find a batch of hacks astir Apple’s walled garden,” he told TechCrunch.

George near Google X successful 2020 and sewage nan thought for TwinMind successful 2023 erstwhile was moving astatine JPMorgan arsenic Vice President and Applied AI Lead, attending back-to-back meetings each day. To prevention time, he built a book that captured audio, transcribed it connected his iPad, and fed it into ChatGPT — which began to understand his projects and moreover make usable code. Impressed by nan results, he shared it pinch friends and posted astir it connected Blind, wherever others showed liking but did not want thing moving connected their activity laptops. That led him to build an app that could tally connected a individual phone, softly listening during meetings to stitchery useful context.

In summation to nan mobile app, TwinMind offers a Chrome extension that gathers further discourse done browser activity. Using imagination AI, it tin visually scan unfastened tabs and construe contented from various platforms, including email, Slack, and Notion.

Techcrunch event

San Francisco | October 27-29, 2025

The startup moreover utilized nan hold itself to shortlist interns from complete 850 applications they received this summer.

“We opened each nan LinkedIn profiles and CVs of nan 854 applicants successful browser tabs, past asked nan Chrome hold to rank nan champion candidates,” George said. “It did a awesome occupation — that’s really we hired our last 4 interns.”

He noted that existent AI chatbots — including OpenAI’s ChatGPT and Anthropic’s Claude — cannot easy process hundreds of documents aliases parse sign-ups from devices for illustration LinkedIn aliases Gmail to stitchery contextual information. Similarly, AI-powered browsers specified arsenic those from Perplexity and The Browser Company deficiency nan expertise to build knowledge from your offline conversations and in-person meetings.

The startup presently has complete 30,000 users, pinch astir 15,000 of them progressive each month. As overmuch arsenic 20–30% of TwinMind users besides usage nan Chrome extension, George said.

While nan U.S. is nan largest guidelines for TwinMind truthful far, nan startup is besides seeing traction from India, Brazil, nan Philippines, Ethiopia, Kenya, and Europe.

TwinMind targets nan wide audience, though 50–60% of its users are presently professionals, astir 25% are students, and nan remaining 20–25% are individuals utilizing it for individual purposes.

George told TechCrunch that his begetter is among nan individuals utilizing TwinMind to constitute their autobiography.

One of AI’s important drawbacks is its imaginable to discuss personification privacy. But George asserted that TwinMind does not train its models connected personification information and is designed to activity without sending recordings to nan cloud. Unlike galore different AI note-taking apps, TwinMind does not fto users entree audio recordings later — nan audio is deleted connected nan alert — while only nan transcribed matter is stored locally successful nan app, he noted.

Google X acquisition helped velocity things up

The TwinMind co-founders spent a fewer years moving connected various projects astatine Google X. George told TechCrunch that he worked connected six projects alone, including iyO — nan squad down AI-powered earbuds, which recently made headlines for suing OpenAI and Jony Ive. That acquisition helped nan TwinMind squad move quickly from conception to product.

“Google X was really nan cleanable spot to hole for starting your ain company,” said George. “There are astir 30 to 40 startup-like projects happening astatine immoderate fixed time. Nobody other gets to activity astatine six early-stage startups complete 2 aliases 3 years earlier launching their ain — astatine slightest not successful specified a short span.”

Before joining Google, George worked connected applying heavy learning to gravitational activity astrophysics arsenic portion of nan Nobel Prize–winning LIGO group astatine nan University of Illinois’ National Center for Supercomputing Applications. He had completed his PhD successful AI for astrophysics successful conscionable 1 twelvemonth — astatine nan property of 24 — a feat that led him to subordinate Stephen Wolfram’s investigation laboratory successful 2017 arsenic a heavy learning and AI researcher.

That early relationship pinch Wolfram came afloat circle years later — he ended up penning nan first cheque for TwinMind, marking his first-ever finance successful a startup. The caller seed information was led by Streamlined Ventures, pinch information from Sequoia Capital and different investors, including Wolfram. The information values TwinMind astatine $60 cardinal post-money.

TwinMind Ear-3 model

In summation to its apps and browser extension, TwinMind has besides introduced nan TwinMind Ear-3 model, a successor to its existing Ear-2, which supports complete 140 languages worldwide and has a connection correction complaint of 5.26%, nan startup said. The caller exemplary tin besides admit different speakers successful a speech and has a speaker diarization correction complaint of 3.8%.

The caller AI exemplary is simply a fine-tuned blend of respective open-source models, trained connected a curated group of human-annotated net information — including podcasts, videos, and movies.

“We recovered that nan much languages you support, nan amended nan exemplary gets astatine knowing accents and location dialects because it’s training connected a broader scope of speakers,” George said.

The exemplary costs $0.23/ hr and will beryllium disposable done an API to developers and enterprises complete nan adjacent fewer weeks.

The Ear-3, dissimilar nan Ear-2, does not support a complete offline experience, arsenic it is larger successful size and runs connected nan cloud. However, nan app automatically switches to Ear-2 if nan net goes distant and past moves backmost to Ear-3 erstwhile it is back, George said.

With nan preamble of nan Ear-3, TwinMind now offers a Pro subscription astatine $15/month, pinch a larger discourse model of up to 2 cardinal tokens and email support wrong 24 hours. Nonetheless, nan free type still exists pinch each nan existing features including unlimited hours of transcriptions and on-device reside recognition.

The startup presently has a squad of 11 members. It plans to prosecute a fewer designers to heighten its personification acquisition and group up a business improvement squad to waste its API. Furthermore, location are plans to walk immoderate money connected acquiring caller users.