Bayen AI — an AI shopping agent, designed end-to-end with Google Stitch
I'm launching Bayen AI before it's finished — the realtime UX and agentic animations I built with Google Stitch were too fun not to share, and I'd rather show a real, in-progress build than a polished demo.
The story. I found Stitch ~3 months ago and finally built the front end I'd been stuck on. I'm building Bayen solo — infrastructure, backend, matching layer, all of it — so the UI kept slipping. Stitch is what unblocked me.
What Bayen is. An AI shopping agent: you prompt it, it matches and finds the right products across sources — AliExpress, Amazon, Alibaba, and more — so you can find, track, and buy in one place through the sources' official APIs. People first; businesses later. This is a product I'm actually shipping, not a challenge demo — just an early phase.
🛠️ What I built
A complete, on-brand frontend, generated with Stitch during the challenge:
Personalized home — auto-matches to signals (location, currency/language, browsing) with a "matched for you" feed.
The agent — an animated prompt composer: Smart / Deep Smart modes, voice, camera (shop-by-photo), a multi-source toggle, and a / skill palette of source "portals" the agent can call — aliexpress_portal, alibaba_portal, amazon_search, /compare, /track, /under.
Realtime PDP — live price + drop indicator, 30-day sparkline, USD/EUR/MAD switcher, and a Compare Offers table flagging the best live source.
Payment handoff — routes the buyer to the source's official checkout (e.g., Amazon) with affiliate disclosure, plus a Track price popup.
Admin — dashboard, orders, returns.
Design: monochrome zinc + one confident red accent, Poppins (JetBrains Mono for prices), an animated lowercase bayen ai wordmark, full light/dark + responsive.
🔗 Link
https://stitch.withgoogle.com/projects/6579973469378019461
(https://stitch.withgoogle.com/projects/6579973469378019461)✨ How I used Stitch
Stitch generated my entire frontend, not just mockups — via its 2026 Vibe Design workflow:
Multi-screen generation — I described each route in natural language; Stitch produced the whole connected flow (home, agent, PDP, payment, admin).
Design system as text — I fed it a structured spec (tokens, type ramp, components) in the spirit of Stitch's DESIGN.md format, so every screen stayed in the same zinc + red + Poppins language.
Streaming design agent — refining by prompt/voice with live render let me pivot the whole system (flat monochrome → Poppins + red, then rebuild the composer and admin sidebar) in minutes.
Newly launched feature showcased: Vibe Design + multi-screen generation, plus the streaming design agent.
🎞️ Motion, interaction & dynamic UI
Send → generation — the prompt rises into a bubble, the agent streams its "thinking", the reply types in, then product cards stagger in with deal badges popping and prices counting up.
Dynamic composer — sliding Smart/Deep Smart toggle, source dropdown with a number-pop, voice waveform + timer, and a / palette that filters as you type.
Realtime feel — a Live Deal Tracker with pulsing "live" dots and price deltas, plus a price-confirming state in the payment handoff.
🔌 External integrations (in progress)
Headless: a Shopify backend for owned catalog/orders, and official source APIs (AliExpress, Amazon, Alibaba) for matching, live pricing, and the buy handoff — with multi-currency FX.
💬 My feedback on Stitch
Loved: the speed (idea → real screens) that finally unblocked my front end; natural-language multi-screen generation that kept the flow coherent; design-system-as-text (DESIGN.md-style) that made a brand pivot ripple across every screen; real, buildable structure, not just pictures.
Would love next: finer motion-timing/prototype-link control so animated states export as a playable flow; deterministic tokens (lock a hex/spacing scale); better real image fills and reusable components.
Net: Stitch went from "fun to try" to the core of my workflow.
🖼️ Screenshots / recording
Agent match feed, PDP with Compare Offers, personalized home, Admin dashboard — plus a recording of the animated prompt → results flow.
🎥 Walkthrough (bonus)
Video attached, and on Loom (https://www.loom.com/share/85be2bf6a40c41c5aceca35ff8f44c01).
Note: brand names (Amazon, AliExpress, Alibaba) are referenced as planned official-API integrations; product imagery is placeholder/generated.
Announcing upcoming Deals AI matching nearby project.
Snatchly will be a new home for deals with super ease simple User Experience and deep AI & Machine Learning process to understand the buyers and suppliers how deals should work and claim.
https://www.snatchly.app/
Announcing a new verified project built with Arete Avril in the UK: Emotions, a unified coaching and emotional support platform focused on helping mothers, young people, children, and adults better understand, express, and manage their feelings in a healthy and supportive way. The platform is positioned around emotional understanding, growth, and connection without judgement, and also extends into family wellbeing, educational resources, podcast content, and emotional support sessions.
Built with Kajabi, Next.js, Flutter, and a scalable modern stack, I developed the solution as a cross-platform mobile and web experience with strong admin capabilities to manage content, engagement, subscriptions, and customer support flows. The product combines a digital self-serve experience with guided services, including subscriptions, emotional support sessions, booking, educational content, and family-oriented resources.
https://areteapex.com/