Burla demo: Analyzing 1.7M Airbnb photos and 50.7M reviews at scale using CLIP and Claude Haiku Vision
By
jmp1062
Plain bagel done well. Pleasantly substantive.
Summary
A technical demonstration of processing all public Airbnb listings (119 cities, 4 quarterly snapshots) using CLIP to score 1.7M photos for suspicious content, with Claude Haiku Vision double-checking shortlists. Also scored 50.7M reviews and reranked the weirdest 12K. All processing was parallelized on Burla's platform using ~1.7K CPU workers and 20 A100 GPUs.
Key quotes
· 3 pulledWe scored 1.7M photos with CLIP (a model that turns an image into a vector you can compare to a text prompt), shortlisted the most suspicious ones, and had Claude Haiku Vision double-check each shortlist.
We also scored every review and reranked the weirdest 12K with Haiku.
Everything was parallelized on Burla, on a single dynamic cluster that scaled to ~1.7K CPU workers for photo download and CLIP, with 20 A100 GPUs.
You might also wanna read
Fluid Storage: Forkable, Ephemeral Infrastructure for AI Agent Era
Fluid Storage is a new infrastructure solution that reimagines block storage with features like zero-copy forks, true elasticity, and synchr
tigerdata.com·7mo ago
Amazon claims networking breakthrough with quasi-random data center design, boosting speed and efficiency
Amazon claims a major breakthrough in data center networking using a "quasi-random" design that combines structured and random network archi
wired.com·1h ago
Google enters AI agent runtime race as the infrastructure layer becomes commoditized
Google repositioned Antigravity as a platform for developing and managing teams of autonomous AI agents at its I/O conference. The platform
bit.ly·8h ago
Amazon Textract achieves PCI DSS certification with enhanced table and form data extraction
Amazon Web Services announces that Amazon Textract, its AI-powered document text and data extraction service, has achieved PCI DSS certifica
aws.amazon.com·16h ago
Snowflake commits $6B to AWS for AI and Graviton compute infrastructure
Snowflake has committed $6 billion over five years to Amazon Web Services (AWS) for Graviton compute and GPU-accelerated EC2 instances, mark
bit.ly·22h ago
AWS Budgets' Multi-Hour Delay Fails to Catch Rapid Bedrock Cost Spikes
This article exposes a critical flaw in AWS Budgets: it can have up to an 8-hour (or more) delay in alerting users about AWS Bedrock spendin
blog.llmcap.io·1d ago