01 - Real Estate Media
An AI-powered real estate news platform built entirely solo. A pipeline of agents scrapes curated Danish sources, filters for relevance, rewrites the content with an LLM, and generates a matching image - then adds it as a draft for a quick human review before it goes live. Built as both a commercial product and my bachelor project.
Sole developer. Designed and built the entire platform - architecture, data ingestion, AI pipeline, editorial workflow, the public-facing site, and ongoing maintenance.
Multi-model pipeline using Gemini and Claude depending on the task. Custom CMS for managing and reviewing articles. Background jobs via Trigger.dev. Free sign-up required to read full articles. Around 200 users and several hundred articles published to date.
The brief was a real estate news platform - and an open question of how far AI could be pushed to fill that need. Real estate news moves daily and a one-person team can't match the volume the market expects, so the real challenge was handing as much of the work as possible to AI while keeping the output accurate, relevant, and trustworthy.
The source pages are all very different, so the scraper had to handle a lot of different situations. Dates were a particular headache - formats and placement varied from site to site - and some pages were static while others were JavaScript-rendered, so the scraper had to reliably handle both.
Trigger.dev background jobs scrape a curated list of Danish sources on a schedule. Each piece of content is analysed for freshness and relevance, then handed to the models: Gemini analyses the content and generates a matching image, and Claude writes the article. Rather than going straight live, each finished piece is added as a draft for a quick review before publishing.
Free registration gates full article access. Mailgun handles transactional and retention emails. The whole product, from pipeline to public site, was designed and built solo.
The pipeline has produced several hundred articles, handling sourcing, writing, and imagery on its own. Each one lands as a draft rather than publishing blind, so the only human step left is a quick review before it goes live.
EstateNews started as a real product concept and was later adopted as my bachelor project at KEA, the idea coming first and the academic framing second. It also doubles as a marketing channel, putting the company's other products in front of a relevant real estate audience.
