Märkische Projekt Bau
How an outdated website became a modern web platform for Märkische Projekt Bau, and then grew into an internal AI portal that converts tender documents into GAEB files and calculates quantity takeoffs from construction plans.
- Industry
- Construction
- Location
- Wildau, Germany
- Duration
- 3 weeks + ongoing
- Live Project
- maerkischeprojekt.de
- Lighthouse score
- 95
- Page load time
- <1.5 s
- on Google for "Bauunternehmen Wildau"
- #1
- Tender to GAEB D83 file
- 1 h → 4 min
A tender PDF goes in, a GAEB file comes out
Tenders used to be typed by hand into the calculation software. In the portal the estimator uploads the PDF and gets a GAEB D83 file to import. In a workshop test an 11-page tender became a valid file in about 3:30 minutes, a job that takes about an hour by hand. The estimator still checks every position before it is used.
The Challenge
Märkische Projekt Bau GmbH, a general contractor established in 1999, had an outdated website that didn't reflect their track record of delivering complex construction projects across Berlin and Brandenburg. Beyond the web presence, their quoting workflow was a bottleneck: tenders arriving as Word or PDF files had to be hand-typed into their calculation software, and quantity takeoffs from construction plans were done entirely by hand, roughly an hour of manual work per document, all resting on a single estimator.
The Solution
The relaunched Next.js site ships in German and English with a dynamic project gallery, a news section, funding guidance, regional landing pages for local search, and an AI chatbot answering visitor queries through the Vercel AI Gateway. Behind a login on the same site sits an internal AI portal with two agents: one converts tender PDFs into valid GAEB D83 files, the other produces quantity takeoffs from uploaded plan PDFs, with selectable frontier models like GPT-5.6 and Claude Sonnet 5, per-run cost tracking, and results that survive closing the tab. Four weeks after the relaunch, the site ranks first on Google for "Bauunternehmen Wildau".
What was built
- Website in German and English with project gallery, news and funding guidance
- Regional landing pages for local search
- AI assistant that answers visitor questions on the site
- Password-protected internal portal on the same site
- Agent that converts tender PDFs into GAEB D83 files
- Agent that produces quantity takeoffs from plan PDFs for the estimator to check
- First place on Google for "Bauunternehmen Wildau" (construction company Wildau), four weeks after the relaunch
Sebastian built and continues to manage our website, providing reliable support for several years. His consistent quality and dependability have been essential for our online presence.
Sebastian KundochManaging Director, Märkische Projekt BauTech Stack
Next.js, React, TypeScript, Tailwind CSS, AI SDK, Vercel AI Gateway, Drizzle, Better-Auth, Resend, Vercel
More Case Studies
Screencurve: an app that watches an episode and writes its format bible
Screencurve analyses film and TV material and writes the format bible for an episode or a series. Every finding links to the timecode and frame it came from, so a producer can check it in the player.
Financing Certificate Platform for a Berlin Real Estate Service
Full-stack platform with automated PDF certificates, 10-step financing form, and admin dashboard for finanzierungszertifikat.com, a real estate service helping property buyers prove their creditworthiness to agents and sellers in under 24 hours.
A Claude Code Research Studio for a Documentary Producer
A private AI research and writing studio for a documentary producer pitching to Germany's largest public broadcasters. Hieronymus Vault combines an Obsidian-native knowledge base, a versioned stylometric corpus, custom Claude Code skills, and a multi-agent fan-out architecture. Idea to broadcaster-ready DOCX in under three hours. In his voice. Every claim source-traced.