Building a mobile bookmarking app with AI shopping recommendations
A bookmarking app for shoppers who save now and decide later, with AI surfacing what's worth coming back to.
The challenge
Shopping decisions don't happen in a single session. A shopper sees a product they like at lunch, gets distracted, and never returns to it. Browser bookmarks don't survive across devices. Wishlist features are siloed per retailer. Screenshots get lost in a camera roll. The thing the shopper wanted to remember disappears into the day.
Twistag built Shoplater as an internal product to solve this category of problem: a single mobile inbox where any product, from any retailer, can be saved with one tap and surfaced later when the moment is right. The challenge was twofold — build a fast, reliable cross-platform mobile experience, and make the recommendation layer smart enough that returning to the saved items felt like discovery, not housekeeping.
What we learned
| Wishlists fragment by store | Every retailer maintains its own list — users default to screenshots instead of a unified view. |
| Per-store integration won't scale | Negotiating API access with thousands of retailers is impossible — the product has to work without their cooperation. |
| Cross-store discovery doesn't happen | Without a unified system, users never find the matching item on a platform they weren't planning to visit. |
The solution
Shoplater is built on React Native, which let us ship iOS and Android from a single codebase without compromising on native feel. The save flow lives in a share extension on both platforms, so a user browsing any retailer's app or website can add a product to Shoplater in one tap. We built the parsing layer in Python on AWS, capable of extracting product details (name, price, image, retailer, availability) from a wide range of source URLs and formats.
The recommendation engine is where the product earned its name. Twistag designed it to surface saved items contextually: items that have dropped in price, items that match user behaviour patterns (more interest in casual wear before the weekend), items the user has saved repeatedly across categories. The recommendations don't just remind users what they saved — they suggest when to act on it.
Cross-device sync was non-negotiable. A shopper who saves an item on their phone should see it on their tablet seamlessly. We built sync on top of AWS infrastructure with conflict resolution at the item level, so user actions across devices stay coherent even with intermittent connectivity.
What this shaped
| Universal scrape beats partner deals | Extract product metadata from any page rather than waiting for partnerships that may never come. |
| Metadata fallbacks survive layouts | When layout changes, Open Graph tags provide a backup path — scraping survives the first redesign. |
| Collections turn saving into planning | Letting users group items into contexts — gifts, mood boards, registries — converts passive saving into active intent. |
The impact
Shoplater works as the consumer-facing testbed for Twistag's product engineering and AI recommendation capabilities. The app demonstrates the patterns we apply to client work: cross-platform mobile delivery from a single codebase, AI recommendations that feel useful rather than intrusive, and the discipline of treating sync and offline behaviour as core product features rather than afterthoughts.
Internal products like Shoplater serve a specific purpose for Twistag: they let us prove patterns at the consumer-app quality bar before applying them to client engagements. The lessons learned shaping the recommendation engine and the cross-device sync architecture have flowed directly into multiple client projects in retail and DTC.
What this proved
| Behaviour reveals product fit | When users abandon screenshots for the app, you have evidence the surface is genuinely better, not just newer. |
| Saves are deeper than visits | A million saves in six weeks isn't traffic — it's evidence the product earns the return. |
| Universality is the moat | Working everywhere without negotiating individual deals stays ahead of platforms tied to specific partnerships. |
Technologies used
- React Native
- Python
- AWS
