I scraped 1.94M Airbnb photos for opium dens, pet cameos, and messy kitchens
burla-cloud.github.io
April 30, 2026
3 min read
🔥🔥🔥🔥🔥
46/100
Summary
Burla analyzed all public Airbnb listings across 119 cities, processing 1.7 million photos using CLIP to identify suspicious images. The review data was scored and reranked, with the entire operation parallelized on a dynamic cluster utilizing approximately 1,700 CPU workers and 20 A10 GPUs.
Key Takeaways
- Burla processed 1.7 million Airbnb photos using CLIP and Claude Haiku Vision to identify and validate suspicious listings.
- The system utilized a dynamic cluster that scaled to approximately 1,700 CPU workers and 20 A100 GPUs for parallel processing tasks.
- A three-tier funnel approach was implemented to filter reviews and photos, allowing users to search and categorize listings effectively.
- Burla enables quick iteration for data teams by allowing Python functions to run across a cluster without the need for Docker or Kubernetes.
Community Sentiment
Positives
- The project effectively stress tests agents and distributed computing, showcasing the capabilities of modern AI infrastructure in handling large datasets.
- Utilizing 1.94 million photos demonstrates the potential of AI to analyze and extract insights from vast amounts of unstructured data, pushing the boundaries of current methodologies.
Concerns
- The exercise raises concerns about the environmental footprint and resource waste involved in scraping and processing such a large dataset without clear value.
- Critics argue that the focus on pet detection lacks significance, questioning the practical applications and overall value of the findings from this AI analysis.