When your architecture is fragmented, you're not just adding complexity, you're encoding fundamental limitations into your entire stack. Here's why a unified model approach creates cascading benefits far beyond the obvious.

While CLIP revolutionized product representation, user modeling has been an afterthought. We explore why treating users and items symmetrically fails to capture the temporal, uncertain, and compositional nature of shopping intent—and what's needed to bridge this gap.

Explore how Composed Item Retrieval is going to change e-commerce by allowing shoppers to modify product searches with natural language, creating a more intuitive and powerful discovery experience.

Time to let the flavours blend together... For too long, e-commerce discovery has been a messy kitchen of separate tools for text, images, and recommendations. We trace the recipe from plain-toast keywords to today's complex dishes, and show why the future is a unified kitchen that can serve up truly novel shopping experiences.

Don't trust the headlines, CLIP is not multimodal! In this blog posts, we explore the multimodal nature of e-commerce and how it can be leveraged for better user experiences.

In this blog post, we explore the cold start problem in e-commerce recommendations, quantifying missed opportunities with traditional systems and showcasing the potential of zero-shot recommendation approaches.

The way you sort products has a direct impact on your bottom line. This blog dives into a simulation that shows how a more dynamic approach can prevent premature stock-outs and ultimately drive higher cumulative sales.

We explore the interplay of token architecture on product design, and model development. We discuss how tokens can be used to create a more capable, flexible and adaptable AI systems.

We unravel the latest squeeze on training bottlenecks, exploring the impact of data quality and pipeline monitoring on training performance.

We go down the rabbit-hole, exploring the recipes that train transformers at scale. These provide insight into network communication, GPU utilization, and the challenges of common transformer ops.

Understand the recipe and you'll know what comes next! We break down the memory requirements of transformer training and weigh in on some innovations born of these constraints.

Delve into the low-level details of Nvidia GPUs are their interaction with PyTorch execution.

Let's jump into the Experiment Tracking - a guiding light in large-scale model training.

Dive into the technology behind vector databases, one of the key innovations helping drive modern AI.

Find out how Zero-shot Discovery unifies the search and recommendations experience in e-Commerce, and the impact legacy algorithms have on user growth.

What makes this wave different? Here are how economics and hardware are building innovation in AI and Machine Learning that will last.

What makes this wave different? Here are how economics and hardware are building innovation in AI and Machine Learning that will last.

Crunch time in AI research! Our tangy tour through the latest breakthroughs in PeFT, Upcycling and Zero-shot Learning will leave you feeling anything but sour.