AI-Powered Housing Assistant

Automating Apartment Hunting in Berlin's Competitive Market

CLIENT:

Self-initiated project

YEAR:

2025

TECHNOLOGIES:

n8n, NocoDB, OpenAI API, Telegram Bot API, Apify

ROLE:

Full-stack developer, AI prompt engineer, workflow architect

challenge.

Berlin's housing market is one of the most competitive in Europe, with demand far outstripping supply. Finding an apartment involves:

  • Quickly filtering through options based on personal criteria

  • Writing personalized application letters for each potential apartment

  • Monitoring multiple websites daily for new listings

  • Responding rapidly before opportunities disappear

This process typically consumes 2-3 hours daily and involves significant repetitive work. The challenge was to create a system that could automate this entire workflow, from discovery to application.

Project Goals.

Automate data collection from multiple housing websites
Filter listings based on natural language user queries
Generate personalized application letters tailored to specific listings
Create a simple user interface accessible to non-technical users
Build a system that could operate autonomously with minimal supervision


The Solution

I designed a multi-agent AI system that handles the entire apartment hunting process through a conversational interface. The system consists of three main components:

1. Data Collection Layer

  • Daily automated scraping of Gewobag and eBay Kleinanzeigen housing platforms

  • Data normalization pipeline to standardize information from different sources

  • Deduplication system to prevent repeated listings

  • Structured database for efficient storage and retrieval

2. Intelligence Layer

  • Data Analyst Agent: Translates natural language queries into database filters

  • Writer Agent: Generates personalized application letters

  • Memory component: Retains user preferences and conversation history

3. User Interface Layer

  • Telegram bot providing a simple chat interface

  • Conversational UI requiring no technical knowledge to operate

  • Real-time notifications for new matches


Architecture Diagram

Data Collection Layer

  • Daily Scrapers (Gewobag & eBay)

  • Apify Web Scraping Platform

  • Automated Deduplication

Storage Layer

  • NocoDB Database

  • Normalized Schema

Intelligence Layer

  • Data Analyst Agent (powered by OpenAI)

  • Writer Agent (powered by OpenAI)

  • Memory Components

User Interface Layer

  • Telegram Bot

  • Conversational Interface

  • Real-time Notifications


Technical Challenges & Solutions

Challenge 1: Inconsistent Data Formats

Different housing platforms structure their listings in entirely different ways, making normalization difficult.

Solution: Created custom data mappers for each source that extracted and standardized critical fields (price, area, rooms, location) while preserving unique information.

Challenge 2: Natural Language Understanding

Users phrase their housing preferences in diverse and unpredictable ways.

Solution: Developed a specialized Data Analyst Agent with carefully crafted prompts that could translate human language into structured database queries, handling variations in how people express constraints.

Challenge 3: Application Personalization

Generic application letters are ineffective in competitive markets.

Solution: Implemented a two-phase approach where the system first asks users for personal details, then combines this information with listing-specific details to generate highly personalized application letters.

Results

The system successfully automated the entire apartment hunting process:

  • Time Savings: Reduced daily apartment search time from 2-3 hours to just 5-10 minutes

  • Coverage: Monitored multiple platforms simultaneously without additional effort

  • Speed: Generated personalized applications within seconds of listing discovery

  • Quality: Created application letters tailored to both listing details and personal profile

Key Metrics

  • Average of 15-20 new listings processed daily

  • 95% accuracy in understanding natural language search queries

  • 30+ personalized application letters generated

  • 100% automation of the data collection process