An AI operating system for galleries, art

consulting, and art wealth management

firms.


by four friends 

from Brown, Harvard, and Columbia.













Vasari understands an entire galleries backlog of data
to improve collection management,
client advisory, and
operations as a whole using AI agents.


DATA


33 artworks, 30 auction house data records as seed;

scraped data: ~150k artworks,

130k auction records, 15k unique

artist profiles, 150k embeddings, 322 auction houses.



Beta tested with RISD galleries,

for data quality check pipelines.



   
$150,000 angel investment






CTO and Design:



      As the technical founder, I was responsible for building this project

      ground up. I had to learn the necessary tools and tech stack in the process

      of building.







At the same time, I had full ownership over the project’s creative direction.  

This meant thinking not only as an engineer,  
but as a designer—  
user experience,  
visual systems,  
branding, and the small details that shape how a product.  
feels    

Making intentional decisions    
at every level—technical and    
aesthetic, even the contemplation    
of what tool to use for a given situation was the scope of my responsibility.    









Tech Stack:





          Database:

          PostgreSQL,
         
          Redis

          AWS S3

          Scrapy



          AI Models being used:

          Claude Sonnet 4.6 (Anthropic) - Vasari Agent

          OpenAI GPT-4o  

          OpenAI GPT-4o Vision

          OpenAI text-embedding-3-small — for artwork embeddings




          AI Architecture:

          RAG pipeline

          Embeddings stored as JSONB

          AI responses cached in Redis (24h for evaluations, 1h for insights).






FrontEnd:



React,

Tailwind CSS,

Framer Motion,

Lucide React.



Deployment: Vercel







BackEnd:



FastAPI,

SQLAlchemy,

Alembic,

Pydantic,

Celery,

WeasyPrint + matplotlib

python-jose (JWT)



Deployment: Railway