Source: Dev.to

SwissContract.ai – From Weekend Hack to Production‑Ready Tool

I shipped SwissContract.ai in a weekend. Told the story. Thought I was done. Then Monday happened.

---

The three cracks that showed up

These weren’t nice‑to‑haves; they were blocking real use.

---

1️⃣ Handling scanned PDFs

Most contracts uploaded are digital PDFs (copy‑pasteable text). A significant chunk, however, are image‑only PDFs.

My original text‑extraction library (unpdf) returns an empty string for those files, causing the “too short” error.

First attempt – OCR in Node.js

• Render PDF pages to a canvas with pdfjs-dist.
• Use OCR on the canvas output.

Problem: OffscreenCanvas isn’t available in Node 22 or on Vercel, so the approach that works in a browser dies on a serverless backend.

The solution – Claude’s native PDF support

Claude can accept a raw PDF (base64‑encoded) and perform text extraction internally. No extra canvas, no OCR library, no server‑side dependencies.

// When text extraction fails ( pathname === `/${l}` || pathname.startsWith(`/${l}/`)
  );
  if (pathLocale) {
    const response = NextResponse.next();
    response.cookies.set("locale", pathLocale, { path: "/", maxAge: 31536000 });
    return response;
  }

  // 2️⃣ Root path only → try cookie, then Accept‑Language header
  if (pathname === "/") {
    const cookieLocale = request.cookies.get("locale")?.value;
    if (cookieLocale && cookieLocale !== "en") {
      return NextResponse.redirect(new URL(`/${cookieLocale}`, request.url));
    }

    const detected = detectLocale(request.headers.get("accept-language"));
    if (detected && detected !== "en") {
      const response = NextResponse.redirect(
        new URL(`/${detected}`, request.url)
      );
      response.cookies.set("locale", detected, { path: "/", maxAge: 31536000 });
      return response;
    }
  }

  // 3️⃣ No special handling → continue as‑is
  return NextResponse.next();
}

Result

• Each language now has a stable, crawlable URL.
• Google discovers /de, /fr, /it as separate pages → proper hreflang handling.
• Shared links always point to the intended language.
• Analytics can segment by pathname.

---

🎉 Takeaways

1. Leverage existing AI capabilities – Claude’s PDF handling saved us from building a custom OCR pipeline.
2. Design i18n with SEO in mind – Locale‑specific URLs are essential for discoverability and analytics.
3. Middleware can keep the user experience smooth while still serving the correct language version.

SwissContract.ai went from a weekend prototype to a production‑ready, multilingual, SEO‑friendly contract analyser. 🚀

Overview

Problem:

• Cookie‑based locale handling broke SEO, link sharing, and analytics.
• The site shipped without essential SEO fundamentals.

Lesson:

• Path‑based routing is the right default for public‑facing sites.
• Ship SEO basics before announcing a product.

---

What Went Wrong

---

Fixes Implemented (in one session)

1. Open Graph Image

• Added an SVG (/public/og-image.svg) – 1200 × 630, fully vector, no rasterisation needed.
• Served directly by Next.js.

2. Sitemap (Next.js 14)

// app/sitemap.ts
import type { MetadataRoute } from 'next';

export default function sitemap(): MetadataRoute.Sitemap {
  return [
    {
      url: "https://swisscontract.ai",
      lastModified: new Date(),
      changeFrequency: "weekly",
      priority: 1,
    },
  ];
}

3. Structured Data (JSON‑LD)

Added FAQPage and Organization schemas to layout.tsx.
Google now can surface FAQ rich results in SERPs.

4. Homepage Content

• Added a FAQ section and a “How it works” step‑by‑step guide.
• These appear only when no analysis is active, keeping the UI clean.

5. Locale Routing

• Switched from cookie‑based locale to path‑based (/de, /fr, …).
• SEO‑friendly and works as the source of truth for language.

---

Product Summary

+-------------------------------------------------+
|                swisscontract.ai                 |
|                                                 |
|  Input:  PDF / DOCX / DOC / TXT                 |
|          (text‑based OR scanned)                |
|                                                 |
|  Language: EN / DE / FR / IT                   |
|            (path‑based, SEO‑friendly)           |
|                                                 |
|  Output: Plain‑language AI analysis              |
|          in the user's language                |
|                                                 |
|  Limits: 5 MB, 20 pages, 5 IP/day               |
|  Storage: none                                  |
+-------------------------------------------------+

• Scanned PDF support – the AI model reads PDFs natively; the simplest solution turned out to be the most powerful.

---

Takeaway

“Ship SEO basics before you tell anyone about the product.”

Skipping those fundamentals cost early traffic and required a painful retro‑fit. The three focused sessions that followed turned a weekend prototype into a usable, SEO‑friendly service.

Live demo: https://swisscontract.ai – analyse any Swiss contract for free, no account required.