Albatross AI Secures $12.5M to Fix Stale Web Recommendations

The internet thinks it knows you, but it’s usually a day late and a dollar short. The product recommendations you see are often based on a search from last week, a click from yesterday, or worse, a purchase you’ve already made. A Zurich-based startup founded by former Amazon AI leaders, Albatross, just raised $12.5 million to fix this, arguing that the web’s discovery problem stems from treating users like static, historical profiles.

The new funding round was led by MMC Ventures, with participation from previous investors Redalpine and Daphni. This brings the Albatross AI funding total to $16 million, following a $3.5 million seed round in 2024. The company is betting that real-time adaptation is the missing piece in the AI puzzle. While most of the industry chases generative AI that can create content, Albatross is focused on what it calls the “second pillar”: AI that understands how you perceive and interact with that content, moment-to-moment.

“Everyone has felt the frustration of seeing the same generic recommendations over and over,” said Dr. Kevin Kahn, co-founder and CEO of Albatross, in a statement. “Our system perceives and adapts instantly, so every search and feed reflects the user’s intent at that very moment.”

How Albatross works in real time

Traditional recommendation engines, even those at major tech companies, typically rely on batch-trained models. They look backward at user history, product popularity, or item similarity, crunch the numbers overnight, and serve up predictions based on old data. Albatross claims to be the first platform that can adapt instantly to changes in user behavior without any manual retraining.

Its platform is built on a transformer-based architecture, the same foundational technology behind models like GPT-4, but is trained directly on live event streams. As a user clicks, scrolls, and searches, Albatross’s sequential embedding models learn from this behavior in milliseconds. The result is a system that doesn’t just know you were looking for running shoes yesterday; it understands that right now, you’ve pivoted to searching for waterproof hiking boots for a trip this weekend.

The company already has two core products: a Real-Time Discovery Feed that dynamically curates content, and a Multimodal Search engine that refines results as a user’s intent evolves. It’s already processing billions of events and tens of millions of predictions monthly for clients in retail, travel, and online marketplaces. According to Albatross, early pilots have shown triple-digit uplifts in user engagement and product discovery, with a full deployment taking less than seven weeks.

For investors, the move represents a necessary evolution. “Personalization is entering a new era,” said Mina Samaan, General Partner at MMC Ventures. “Albatross moves beyond static algorithms to build systems that understand context as it happens.”

The irony of ex-Amazon AI leaders tackling this problem is not lost. Amazon pioneered online personalization, but its system can still feel clumsy, recommending items you just bought or things tangentially related to a one-off gift purchase. The Albatross team, which also includes co-founders Dr. Matteo Ruffini and serial entrepreneur Johan Boissard, is essentially building the system they believe should have come next.

As Redalpine’s Dr. Marc Moesser put it, the team has “gone beyond what even the largest platforms achieve with real-time infrastructure.” In an online world saturated with content and products, the biggest challenge is no longer just making things, but helping people find them. Albatross is betting its real-time brain can finally close that discovery gap.