Launching today
Poshuk
AI product search for e-commerce that actually understands
5 followers
AI product search for e-commerce that actually understands
5 followers
Poshuk is an AI-powered product search and discovery engine for e-commerce. Instead of keyword matching, it uses semantic search to understand what shoppers actually want - matching intent, not just strings. Store owners get better on-site search that surfaces the right products faster, reducing bounce and lifting conversion. Built for SMBs who can't afford enterprise search but need results that feel like it.
Hey Product Hunt! 👋
I'm Daniil, founder and developer behind Poshuk (Belarus released).
I built this because on-site search on most stores is genuinely broken. Type "gift for a 5-year-old who likes dinosaurs" and you get zero results - because keyword matching only knows the words, not the meaning. Meanwhile the shopper leaves and buys elsewhere.
Poshuk fixes that with semantic search. It understands intent, so shoppers find the right product even when they don't know the exact term for it. We're already aggregating 1.5M+ products, and the difference in how people find things is night and day.
It's built for small and mid-sized stores that can't drop enterprise money on search but still lose sales every day because of it.
I'd love your feedback - especially from anyone running an online store:
What's the most frustrating search experience you've had as a shopper?
Store owners: how much do you actually trust your current search?
Happy to answer anything about the tech, the model, or the roadmap. 🙏
How does Poshuk handle the cold start problem for new stores that don't have much product data or search history to train the semantic model on?
@cemalfzvr There's no per-store model that needs training, so the cold-start problem mostly doesn't apply. Search works from day one on the catalog text itself, pretrained embedding model turns products into vectors instantly, plain keyword search matches names and descriptions, and an LLM handles understanding the query and ranking results. Per-store tuning (category synonyms, capabilities) is generated by an LLM from the store's own catalog, not from click or search history. The only setup step is a one-time "generate embeddings" run, until then the store still works on keyword search.