Our client is a renowned publishing company that have been successfully providing domain specific, high-quality content for dedicated target groups for decades. Since high-quality content has experienced devaluation due to the unlimited amount of free, commoditized content accessible across the internet, the top management decided to discover new business and growth opportunities by combining state-of-the-art AI technology with their existing products and assets. Together, we started with nothing but a courageous target to not just monetize data but to build entirely new customer experiences that eventually comprise a future-proof and profitable product and business.


The valuation process for real estate is heavily regulated in Germany involving many manual, bureaucratic, and time-consuming tasks. To achieve tremendous gains in efficiency along the entire journey from visiting properties to the generation of the final appraisal, we build a full-stack application for multiple devices covering the daily workflow end-to-end. Our attention focused focused on two main success factors:

  1. Creating a unique and “natural” user experience between humans and AI enabling. Both seamless interaction and controllability by keeping humans in the loop had high priority.
  2. Training and orchestrating multi-modal AI models comprising the core of our product. These models take care of many complex tasks from analyzing data, extracting information from audio, video and images, reasoning over unstructured text, summarization, clustering and transferring data into actual real estate appraisals.

More information about the solution can be found here

AI Tech Stack

  • OpenAI Large Language Models (early stage)
  • Fine-tuned open-source LLMs (later stage)
  • Azure OpenAI model deployments within GDPR region
  • AWS Sagemaker for deployment and inference of open-source models
  • Llava 1.6 for multi modal features
  • Whisper for audio processing

Our Approach

Since we only had a strategic objective and its associated constraints, we had to identify problems worth solving large enough to build a future-proof business upon.

In the following, we describe our battle-tested methodology and create solution ideas that are desirable for target audiences, viable for the business and feasible to be built.


The first eight weeks of the project were about to

  • understand the market and its frictions
  • reveal the pains, needs and desires of real estate appraisers
  • ideate multiple solutions and create design concepts
  • conduct feasibility studies by rapid prototyping

At the end of the phase we had a portfolio of ideas potentially adding significant value to the job of real estate appraisers. Each of the idea included AI as the core technology from a different perspective.


Before we start building a solution we test and validate by applying different technology-agnostic techniques and strategies to gather quantitative and qualitative market data. With this approach we collect as much data as possible to evaluate the risk-reward-ratio of an investment into AI. For this specific project we tested online ad campaigns, email campaigns, newsletters and offline marketing channels to get feedback for our value proposition. After four to weeks of testing we had a solid data base backing the decision making process. Moreover, we had many signed LOIs without having started the actual development of the product. When building a new business model is part of the project, we extend the prototyping phase with before mentioned validation tactics for sake of viability testing.


After twelve weeks of finding and testing the right solution, we started to build the AI application. With a team of fullstack developers, AI engineers, user interface and experience designers, product owner and venture architect, we composed an interdisciplinary SCRUM team that executes in fast and lean two week sprints over the course of nine months. During that time we follow cycles of building, testing with users and iterating until the final market launch. We are not only collaborating closely with our client but also with a a focus group of appraisers that are constantly giving feedback. This human-centric approach ensures to deliver a final AI solution humans love to use.