My background is in e-commerce and digital marketing, and there's a frustration I've seen from both sides of that world: marketing teams spend real budget driving traffic to a store, and then the store's own search bar loses a huge share of those visitors before they ever find what they came for. A customer types "something for a wedding" or "black loafers under $80" and gets either an empty results page or a wall of irrelevant products, because most on-site search is still just keyword matching from a decade ago. That gap between "we paid to get this person here" and "we couldn't actually help them find anything" always bothered me.
What struck me building this challenge is how solvable that gap actually is without needing an enterprise AI budget. A small merchant can't afford to bolt an LLM onto every search query as the cost adds up fast at real traffic volume. So I wanted to prove that a genuinely adaptive, conversation-aware shopping agent — one that tells the difference between someone browsing and someone ready to buy, and asks the right follow-up question instead of just showing more of the same. It can run on cheap, deterministic retrieval and still meaningfully outperform standard search. That's the version of this I actually want to exist: better product discovery that any store can afford, not just the ones with the biggest budgets.
Built With
- bm25
- information-retrieval
- pytest
- python
- scikit-learn
- scipy
- sqlite
- tf-idf
Log in or sign up for Devpost to join the conversation.