Artizan: teaching a platform to estimate what a piece of art is worth

Art prices are scattered across auction houses, galleries, social feeds and news. Artizan pulls those signals together for artists and artworks, then produces a valuation estimate that someone can read and question.

I own features from data ingestion through to the interface. The work spans the scraping layer, the pipelines that clean and model the data, and the Next.js screens where people read the result.

• Apify scrapers that collect auction records, artist profiles, social media and news.
• Python and Dagster pipelines that normalise that data into Supabase/PostgreSQL.
• Vector and embedding search, so comparable artists and works can be found by meaning, not only by name.
• LLM-driven analysis and agent workflows that turn raw records into a written valuation.

A valuation is only useful if people trust it. Keeping the data lineage clear, from the scraped record to the final estimate, is what makes the output defensible.

• Next.js
• Python
• Dagster
• Supabase
• PostgreSQL
• Apify
• OpenAI
• Embeddings