Inspiration

Imagine losing your sight and losing your independence at the checkout — forced to ask someone else just to know if a jacket is warm enough, or if a cream will irritate your skin. We built ShopSense to give that independence back.

What it does

ShopSense is a voice-first AI shopping assistant that lets visually impaired users ask real questions — "Will this fit me?", "Is it actually waterproof?" — and get clear, screen-reader-optimized answers in seconds. It even catches when a brand's claims don't match what real buyers experienced, so users get the truth, not just the marketing.

How we built it

At the heart of ShopSense is a ReAct agent loop — think of it as an AI that thinks, searches, reflects, and only then answers — pulling from customer reviews, expert knowledge, and visual product descriptions stored in a Qdrant vector database. The frontend is fully WCAG 2.1 AA compliant, with voice input, high-contrast themes, and ARIA live regions that make every interaction accessible by design.

Challenges we ran into

Getting an LLM to consistently speak in the language of a screen reader — short, verdict-first, zero symbols — took far more iteration than we expected. And building conflict detection without any labeled training data meant designing a signal-matching system from scratch that could generalize across topics like warmth, sizing, and waterproofing.

Accomplishments that we're proud of

Every accessibility feature in ShopSense is functional, not decorative — voice input, text-to-speech, high contrast, and screen-reader formatting all work together as one seamless experience. And the Agent Reasoning panel means users can see exactly why the AI said what it said — transparency that builds real trust.

What we learned

Accessibility isn't a layer you add at the end — it shapes every decision, from how you write a system prompt to how you lay out a button. And we learned that a well-structured retrieval pipeline beats a confident hallucinating LLM every single time.

What's next for ShopSense

We're just getting started — next up is multi-language support, real e-commerce integrations, and a native mobile app that works seamlessly with VoiceOver and TalkBack. Because accessible shopping shouldn't be a demo — it should be the default.

Share this project:

Updates