Every click online tells a story. Yet the systems behind most of the internet still treat users as static profiles, recycling yesterday’s data to predict tomorrow’s intent. Albatross, a Zurich-based AI company founded by former Amazon AI leaders, has raised $12.5 million in new funding to rewrite that logic with the world’s first platform for real-time product and content discovery – one that learns, reasons, and adapts as users interact.
The round was led by MMC Ventures with participation from Redalpine, Daphni, and strategic angels, bringing Albatross’s total funding to $16 million, following a $3.5 million foundation round in September 2024 led by Redalpine. The company’s platform is already serving billions of live events and tens of millions of predictions each month across marketplaces, retail, and travel platforms worldwide, processing approximately a hundred million products and tens of millions of end users.
Founded in 2024 by Dr Kevin Kahn and Dr Matteo Ruffini, both former Amazon AI leaders, alongside serial entrepreneur Johan Boissard, Albatross is tackling what the team sees as a fundamental gap in the AI revolution. While much of the industry focuses on large language models that generate content, Albatross is building the second pillar of AI: understanding how users perceive and interact with content in real time. It is built on transformer-based architecture with sequential embedding models trained directly on live events.
Traditional recommendation systems look backward, using batch-trained models that rely on popularity, similarity, or user history. They struggle to capture what really matters: what a person is doing right now. In contrast, Albatross replaces these legacy systems with AI that learns continuously from live behavior, updating in milliseconds as users browse, search, and explore without any manual intervention or retraining. Notably, until now no platform could adapt instantly to changes in user behavior. Albatross is the first to do this.
Albatross’s two flagship products – the Real-Time Discovery Feed and Multimodal Search. The Discovery Feed dynamically curates inspiring products and content in real time, while the Multimodal Search engine refines results based on evolving intent, even bridging in-store and online journeys through contextual and image input. The platform operates with enterprise-grade reliability at virtually zero latency.
Early pilots have shown triple-digit uplifts in engagement and product discovery. Integration takes less than seven weeks from signature to deployment, and the platform operates with enterprise-grade reliability, handling billions of data points. The company’s research on cold-start discovery, presented at RecSys 2025, now powers its production platform at scale.
As content and commerce continue to explode, discovery is becoming the defining challenge of the digital economy. Albatross’s goal is to make digital experiences adaptive – transforming the way people find what inspires them, in real time.
Guest Post: US shopping apps collect more data than Chinese or Canadian rivals
Posted in Commentary with tags Surfshark on November 18, 2025 by itnerdAs shoppers gear up for the holiday season, Surfshark investigated the data collection practices of the 10 most popular shopping apps in the US, finding that US-based apps tend to collect more data compared to their counterparts in China and Canada. For example, Amazon collects 25 unique data types out of 35, but among Chinese apps, Alibaba is the most data-hungry, collecting 19 unique data types.
“Scrolling through tempting deals on Temu, Shein, Amazon, and other shopping apps is a Black Friday tradition for many. However, before downloading any shopping app, people should consider whether they are truly willing to trade their privacy for a discount,” says Miguel Fornes, Information Security Manager at Surfshark. “Many shopping apps collect far more data than people realize, and this extends beyond purchase history. Some apps can even gather sensitive information such as political views, racial background, or biometric and health data.”
The Amazon shopping app is the most privacy-intrusive. It collects 25 unique data types out of 35, Walmart and Costco each collect 23, and Whatnot — another US-based app — collects 20. Among Chinese apps, Alibaba is the most data-hungry, collecting 19 unique data types, followed by Temu with 17, Aliexpress with 16, and Shein with 15. The Canadian app, Shop, collects 19 data types, which places it on par with the most data-collecting Chinese app.
All the analyzed apps collect information such as email address, name, payment information, physical address, user ID, search history, and product interaction. The majority of these apps also gather device IDs (except for Temu), phone numbers (except for Shein), photos or videos (except for Shop), and location data (except for Shein). Additionally, most of this collected data is directly linked to individual users, enabling these apps to build comprehensive user profiles, which raises privacy concerns.
Some of the data collected by these shopping apps is surprising and even bizarre. For instance, Amazon and Walmart collect sensitive information — which could include political opinions, racial or ethnic background, biometric data, genetic information, sexual orientation, disability status, or pregnancy details. Whatnot and Alibaba collect users’ contacts, such as contact lists from a user’s phone or address book. In addition, Amazon, Walmart, Whatnot, and Alibaba collect users’ voice or sound recordings.
According to Fornes, these abusive data collection practices can be very dangerous if an app is breached and information about a person is leaked. First, leaked bank account information and purchase history can lead to unauthorized charges, identity theft, and significant financial loss. Second, leaked sensitive information – especially sensitive data like political views or health data – can damage your reputation and financial standing, as health data rarely changes and may be used by insurance and healthcare companies. Finally, all this leaked data might fuel subsequent highly personalized phishing campaigns. Therefore, Fornes advises:
For the complete research material behind this study, visit here.
Leave a comment »