Ex-Google Team Raises $6M for AI 'Second Brain' - Trio AI

A New AI Aims to Function as Your "Second Brain"

A team of three former scientists from Google X is developing an AI-powered application designed to act as a virtual extension of your memory. This isn't about futuristic implants, but rather an app that learns through continuous listening, capturing the context of your daily life.

TwinMind: The Startup Behind the Innovation

TwinMind, the startup spearheading this project, has successfully secured $5.7 million in seed funding. They have launched both Android and iPhone versions of their application, alongside a newly developed AI speech model.

How TwinMind Captures and Processes Information

Founded in March 2024 by Daniel George (CEO), Sunny Tang, and Mahi Karim (both CTOs), TwinMind operates discreetly in the background.

With user consent, it records ambient speech to construct a personalized knowledge graph. This allows the app to transform spoken thoughts, meetings, and discussions into organized information.

Key Features and Capabilities

• AI-Powered Notes & To-Dos: Generates actionable insights from captured audio.
• Offline Functionality: Processes audio in real-time directly on your device.
• Extended Battery Life: Can record continuously for 16-17 hours on a single charge.
• Data Backup: Offers optional data backup for conversation recovery.
• Real-Time Translation: Supports translation across more than 100 languages.

Differentiating from Existing AI Note-Taking Solutions

TwinMind distinguishes itself from competitors like Otter, Granola, and Fireflies by passively recording audio throughout the day. The team achieved this by creating a low-level service in Swift specifically for iPhones.

This approach bypasses the limitations imposed on cloud-based processing by Apple, which restricts prolonged background operation for many applications, as explained by George in a recent interview.

The Genesis of TwinMind

Daniel George departed from Google X in 2020. The concept for TwinMind emerged in 2023 while he was serving as Vice President and Applied AI Lead at JPMorgan.

Facing a demanding schedule of back-to-back meetings, he developed a script to capture audio, transcribe it, and feed it into ChatGPT. The AI’s ability to understand his projects and even generate code impressed him, leading to the development of TwinMind.

Expanding Context with a Chrome Extension

Beyond the mobile app, TwinMind provides a Chrome extension that enhances contextual understanding through browser activity. Utilizing vision AI, it scans open tabs and interprets content from platforms like email, Slack, and Notion.

The extension was even utilized internally to streamline the intern selection process this summer.

Streamlined Intern Hiring Process

“We leveraged the extension to rank candidates from over 850 applications,” George stated. “It performed remarkably well, and that’s how we identified and hired our four interns.”

Addressing Limitations of Current AI Technologies

Current AI chatbots, including ChatGPT and Claude, struggle with processing large volumes of documents or extracting contextual information from sources like LinkedIn and Gmail. Similarly, AI-powered browsers lack the ability to learn from offline conversations and in-person interactions.

User Growth and Demographics

TwinMind currently boasts over 30,000 users, with approximately 15,000 active monthly users. Around 20% to 30% of users also utilize the Chrome extension.

While the U.S. represents the largest user base, the startup is experiencing growing adoption in India, Brazil, the Philippines, Ethiopia, Kenya, and Europe.

Target Audience

TwinMind appeals to a broad audience, with professionals comprising 50% to 60% of users, students accounting for around 25%, and the remaining 20% to 25% using it for personal applications.

George shared that his own father is utilizing TwinMind to document his life story.

Prioritizing User Privacy

Acknowledging concerns about AI and privacy, George emphasized that TwinMind does not train its models on user data and operates without transmitting recordings to the cloud.

Unlike many AI note-taking apps, TwinMind does not allow users to revisit audio recordings; the audio is deleted immediately after transcription, with only the text being stored locally within the application.

Insights from Google X Accelerated Development

The co-founders of TwinMind dedicated several years to diverse initiatives within Google X. George shared with TechCrunch that he contributed to six distinct projects, notably including iyO – the developers of AI-powered earbuds, recently recognized for their legal action against OpenAI and Jony Ive. This prior experience significantly facilitated the TwinMind team’s rapid progression from initial concept to a tangible product.

“The environment at Google X proved to be ideal preparation for launching a new venture,” George stated. “At any given time, approximately 30 to 40 projects operate with a startup-like dynamic. Few individuals have the opportunity to participate in six early-stage startups over a two to three-year period before initiating their own – particularly within such a condensed timeframe.”

Prior to his tenure at Google, George focused on the application of deep learning techniques to the study of gravitational wave astrophysics, collaborating with the Nobel Prize-winning LIGO group at the National Center for Supercomputing Applications, University of Illinois. He successfully earned his PhD in AI for astrophysics within a single year – at the age of 24 – an accomplishment that subsequently led to his appointment as a deep learning and AI researcher at Stephen Wolfram’s research lab in 2017.

This initial association with Wolfram proved pivotal years later; he became the first investor in TwinMind, representing his inaugural startup investment. The latest seed funding round was spearheaded by Streamlined Ventures, with contributions from Sequoia Capital and other investors, including Wolfram. TwinMind’s post-money valuation following this round is $60 million.

TwinMind Ear-3 Model: Advancements in AI Speech Recognition

TwinMind has expanded its offerings beyond its applications and browser extension with the release of the TwinMind Ear-3 model. This new iteration builds upon the capabilities of the existing Ear-2, providing support for over 140 languages globally.

According to the company, the Ear-3 achieves a word error rate of 5.26%. A key improvement lies in its ability to distinguish between multiple speakers within a single conversation.

Enhanced Speaker Recognition

The Ear-3 model demonstrates improved speaker diarization, meaning it can accurately identify who is speaking when. Its speaker diarization error rate is reported to be 3.8%.

This functionality is achieved through a carefully refined combination of open-source AI models.

Training Data and Methodology

The model was trained using a specifically curated dataset of human-annotated data sourced from the internet. This data includes a diverse range of audio content, such as podcasts, videos, and movies.

George, a representative from TwinMind, explained that supporting a wider array of languages contributes to the model’s overall accuracy. Training on a broader spectrum of speakers improves its understanding of accents and regional dialects.

Availability and Cost

The Ear-3 model is priced at $0.23 per hour of use. Access will be granted to developers and enterprises via an API in the coming weeks.

Cloud-Based Operation and Fallback

Unlike its predecessor, the Ear-2, the Ear-3 model operates entirely in the cloud and does not offer full offline functionality. However, the application is designed to seamlessly revert to the Ear-2 model when an internet connection is unavailable.

Once connectivity is restored, the system automatically switches back to utilizing the Ear-3 model.

Subscription Options

TwinMind has introduced a Pro subscription tier priced at $15 per month. This subscription provides access to a larger context window, capable of processing up to 2 million tokens, and includes 24-hour email support.

A free version remains available, offering all existing features, including unlimited transcription hours and on-device speech recognition.

Future Development and Team Expansion

Currently, the TwinMind team consists of 11 individuals. The company intends to expand its team by hiring designers to improve the user experience and establish a business development team focused on API sales.

Furthermore, resources will be allocated to user acquisition efforts.