Persé

Persé is your personal museum curator, humanizes your favorite pieces from the Harvard Art Museum. It's a digital tool to bridge art provenance, historical context, and the lived experience.

Comment

Here is our demo: https://www.canva.com/design/DAGlL-5Rrl8/bZgyfFr1o8UPWjFHuAGUEQ/view?utm_content=DAGlL-5Rrl8&utm_campaign=designshare&utm_medium=link2&utm_source=uniquelinks&utlId=h9c73875e3a

Our Journey Building Persé, the Museum Curator You Can Talk To

What Inspired Us

We started with a simple but powerful belief: that art deserves to be experienced emotionally, not just academically. While visiting museums, we often found ourselves longing for something more personal than the usual wall plaques, and instead something that could tell the story behind the piece, the people who touched it, and the worlds it has passed through.

That idea gave birth to Persé, a digital curator built during the Harvard WiCS Hackathon. Persé is designed to narrate the hidden journeys of artworks in the Harvard Art Museums using the capabilities of the Claude API by Anthropic.

What We Learned

As first-time builders in the AI space, we were struck by the nuance and emotional intelligence Claude brings to storytelling. We learned how to craft prompts that don’t just ask for information but evoke empathy, curiosity, and imagination. We explored Claude’s strengths, like grounding creative narratives in historical detail, and learned to guide the model without over-constraining it.

We also developed a deeper appreciation for provenance itself, not just as a sequence of ownership, but as a rich canvas for human stories that span continents, revolutions, and generations.

How We Built It

We built Persé as a Flask-based web application using:

  • The Harvard Art Museums API to access object images, metadata, and provenance
  • Anthropic’s Claude API to generate emotionally intelligent, historically grounded narratives
  • A lightweight HTML and CSS front end where users can input an object ID and receive both an image and a story

We designed prompts that blended structured data (artist, medium, culture, provenance) with open-ended requests for reflective storytelling. We iterated on these prompts to find the right balance of tone, detail, and emotional connection.

Challenges We Faced

Getting the system to run smoothly across APIs came with its own challenges. We encountered:

  • Authentication errors with the Claude API that required debugging token issues
  • Unexpected response formats that made parsing Claude’s output more complex than expected
  • Prompt tuning, which involved learning how to ask for creativity while staying historically grounded
  • Token limits and rate caps that constrained how much data we could pass into a single request

Each obstacle gave us a better understanding of how to troubleshoot AI systems and work collaboratively through ambiguity.

What Persé Means to Us

Persé was our first project using large language models, and it changed the way we think about both storytelling and technology. It showed us that AI can be used not just for analysis or automation, but for bridging the emotional and historical distance between people and objects.

We hope Persé helps others experience art not just as observers, but as participants in a longer story that is still unfolding.

Built With

Share this project:

Updates