Kimchi 1: Product Search Enters the Post-Training Era
Post-training is reaching retrieval. Here is the formal construction—Plackett–Luce policy gradients under a frozen rubric judge—and why product search is the right place to build it.
In a sea of search and recommendation providers, Solenya (yes, it means “pickle”) was founded on a belief that e-commerce discovery deserved better.
The discovery ecosystem was stuck in separate jars-search in one, recommendations in another-fermenting in isolation and limiting what experiences could be created.
Most platforms need days of data to pickle their recommendations. Zero-shot means instant, crisp recommendations from day zero-for new users, new products, new partners, and even new domains.
Our multimodal approach puts visual understanding at the core-because in e-commerce, a picture really is worth a thousand SKUs.
The secret to our brine? Solenya leverages large-scale, self-supervised learning to soak up knowledge from massive datasets before ever touching your catalog.
This gives our zero-shot, multimodal models the depth of flavorto understand what users want, even when they're searching for something they've never seen before.
Ready to spiceup your platform? Let's talk about what we're fermenting next.
Open-notebook research. Methods, confidence intervals, and the experiments that didn't work.