About the role
AI Agent for Automated Real Estate Video Understanding & Editing
DeSH Real Estate GmbH is an international real estate brokerage company connecting German real estate with international buyers and investors.
In this Project Study, students will develop and validate a multimodal AI Agent that transforms raw property footage into automatically planned and edited real estate marketing videos.
Today, producing professional property videos requires substantial manual work: editors need to review raw footage, identify usable scenes, select the best clips, define the sequence, trim and edit the material, add property information, perform quality checks, and export videos for different channels.
The goal of this project is to investigate how multimodal AI, computer vision and AI agents can automate major parts of this workflow.
The project covers:
automatic segmentation and analysis of raw property footage
recognition of rooms, property areas, shot types and camera movements
detection of blurred, unstable, redundant or unsuitable footage
intelligent clip ranking and selection based on visual quality, relevance and scene coverage
AI-based storyline and editing-plan generation
creation of a structured, machine-readable editing plan containing clip selection, in/out points, sequence order, duration, speed, cropping, transitions and overlays
programmatic execution of the editing plan in professional video-editing software
AI-assisted visual enhancement where appropriate
automated technical, visual and editorial quality control
generation of editable video timelines rather than only final rendered videos
automated export for vertical 9:16 and horizontal 16:9 formats
A key concept of the project is to separate AI reasoning from editing execution: the AI first understands the available footage and generates an explicit editing plan, which is then executed programmatically in the editing environment.
The final prototype should be tested using representative real-world property footage from DeSH Real Estate.
Potential evaluation criteria include:
scene and room recognition accuracy
clip-selection quality
reduction of redundant or unsuitable footage
quality and coherence of automatically generated edits
percentage of editing operations executed automatically
number of manual corrections required
reduction in manual editing time
Project format: TUM Project Study
Team size: 2–5 students
Duration: 3 months full-time or 6 months part-time
Earliest start: 01 October 2026
Work mode: 100% remote. All project coordination, meetings and feedback sessions can be conducted online. No office attendance is required.
We welcome students interested in AI, multimodal models, computer vision, video processing, automation and real-world technology applications. Prior real estate experience is not required.
How to apply
Please send your CV, transcript of records, and a short motivation statement to the email address above.
If available, please also include links to relevant GitHub projects, portfolios, or previous AI/data/automation projects.
Please use the subject line: TUM Project Study – AI Agent – [Your Name]