🧠 Colle: Visual Thinking for Your Research

Colle is a smart, visual research companion that transforms fragmented paper reading into structured insight. It helps researchers highlight, remix, and generate figures, and build a visual memory of their readings through interactive visual cards. With Colle, you can organize key takeaways, revisit annotated insights, and spark deeper conversations by comparing cards with peers.

Inspiration

As researchers, we read dozens to hundreds of papers, yet struggle to remember what we’ve read, let alone share or build upon it. I found myself scanning through figures, taking scattered notes here and there, and losing track of insights I once had. Tools like Zotero or Notion help with storage, but not with intuitive visual thinking, synthesis, or collaboration.

This project was inspired by the idea of building a "second brain for researchers" — one that leverages visual and interactive thinking to hopefully help us:

Understand more deeply
Remember more easily
Discuss more meaningfully

How we built it

The core components of Colle include:

🧾 PDF Reading + Insights Tracking
Logs users interactions with Gemini when reading papers to personalize visual cards.
📊 Interactive Figure Interface
Users can select, annotate, or remix texts and figures in the paper.
🧠 Mindmap Card Generator
Each paper becomes a dynamic card, combining extracted metadata, user insights, key visuals, and AI-summarized takeaways.

Challenges we ran into

Limited access to online paper sources
We initially hoped users could simply paste a link (e.g., arXiv or publisher URLs) to load any paper directly into the Colle PDF viewer. However, this proved infeasible due to access restrictions (e.g., paywalls), lack of consistent LaTeX sources, and issues with low-resolution or unstructured PDFs. As a result, we currently only support local PDF uploads.
Exploratory interactions vs. concise summaries
Users often engage in exploratory, non-linear reading: highlighting tangents, making speculative annotations, or jumping between sections. While this is natural and valuable, it creates noise when generating visual summaries. One major challenge has been designing intelligent systems that can distill meaningful insights from this exploratory behavior without oversimplifying or overwhelming the user.

What we learned

Visual memory is powerful: when users manipulate or generate visuals themselves, they retain insights longer.
It's crucial to balance structure and flexibility: too rigid and users feel constrained, too open and they lose coherence.

What's next for Colle: A visual memory for every paper you read

Card comparison across researchers
We're working on a feature that allows multiple researchers to compare their own cards for the same paper, surfacing different interpretations, complementary takeaways, or even conflicting emphases. This can spark rich discussion, enable peer learning, and support group-based literature reviews or journal clubs. We believe this kind of multi-perspective synthesis will be a powerful way to deepen understanding and encourage critical dialogue.