001 / website
Green Bear
A faster, bilingual site built to win over mobile visitors near the trail
- Client
- Green Bear
- Industry
- Outdoor adventure tourism
- Year
- 2026
- Location
- Tara National Park, Western Serbia
- Role
- DesignFront-end engineering
- Stack
- Static HTML · Python build pipeline · Vanilla JS (IntersectionObserver) · Self-hosted woff2 · WebP

What Green Bear got
A site that loads almost instantly, even on the weak mobile signal their guests have near the trailheads, works fully in both Serbian and English, and is built to be found by search engines and AI answer tools. Along the way we cut the site's image weight from 135 MB to 14 MB and removed a copyright problem that was a real legal risk before launch.
The client
Green Bear runs guided outdoor trips across Tara National Park and the Drina canyon in western Serbia: stand-up paddleboarding, hiking, off-road tours, snowshoeing, splitboard, and fishing.
The opportunity
The old site worked, but it was quietly costing bookings.
- It loaded slowly. The image library was 135 MB of heavy, unoptimized photos. Green Bear's guests are usually outdoors on a weak mobile connection, exactly where a slow, heavy page makes people give up.
- It looked generic. It used an off-the-shelf template that looked like any other tourism site, when the photography of the Drina canyon should have been the star.
- It felt unfinished. Some galleries still showed placeholder "lorem ipsum" text, which makes a small operator look amateur to a first-time guest.
- It carried a legal risk. The signature "house on the Drina" photo was a copyrighted image the client did not own, and another photo had a competitor's watermark on it. Both had to go before launch.
What we did
We rebuilt the site from scratch as a fast, lightweight site that works in both languages and depends on nothing external to load.
- Speed: we cut the image weight from 135 MB to 14 MB (same look, far lighter) and removed external font loading, so pages load fast even on a poor connection.
- Two real languages: Serbian and English each get their own proper pages (not a switch that swaps text after the page loads), so Google shows the right language and nothing breaks.
- A distinctive look: a custom "Expedition Log" design built for the outdoor market, so Green Bear stands out instead of blending in.
- Real content: we replaced the placeholder text with finished, rights-cleared material, and removed the two photos that were a legal risk.
- A booking enquiry form that works with no server to maintain.
The results
We report only what we measured directly. The old site was never measured, so the table below is the before/after we can verify, and the Lighthouse scores are lab results from the new build.
| Metric | Legacy infrastructure | Numscii build |
|---|---|---|
| Image payload | 135 MB source library (1 to 2 MB/asset, PNG) | 14 MB (WebP, responsive sizing) |
| External font requests | Google Fonts (render-blocking) | 0, self-hosted woff2 (~310 KB) |
| Front-end runtime | Webflow runtime + jQuery | None, hand-written vanilla JS |
| Rendering model | Platform-templated | Pre-rendered static HTML |
| Localization | Limited | 16 locale-specific URLs + hreflang |
| Content integrity | Lorem ipsum in galleries | Finalized, curated, rights-cleared |
| Copyright exposure | 1 third-party photo + 1 watermarked asset | Eliminated |
Runtime performance, Lighthouse, measured 15 June 2026. The legacy site had no baseline to compare against, so the figures below are the new build's Lighthouse lab scores. Real-world (CrUX field) data is not available yet, which is expected for a site that has not accumulated enough traffic, so these are lab results, reported as such.
| Lighthouse category | Desktop | Mobile |
|---|---|---|
| Performance | 100 | 80 |
| Accessibility | 100 | 96 |
| Best Practices | 100 | 100 |
| SEO | 100 | 100 |
Built to be found
Because the full text of every page is in the HTML from the first byte, with no JavaScript needed to read it, search engines and AI answer tools (Google AI Overviews, Perplexity, ChatGPT Search, Gemini) can read and cite Green Bear reliably, in the right language. A structured-data layer describes the business, its location, service area, and activities, so it can be cited accurately rather than guessed at.
Outcome
Green Bear went from a heavy, template-bound, partly-unfinished site with a legal exposure to a fast, distinctive, rights-clean site that works in both languages and can be hosted anywhere with no subscription, no backend, and nothing external to break.
Want results like these?
Tell us about your project. We will respond directly, no sales queue.