Overview

Rabbit Hole is an AI-powered research exploration platform I developed to enhance academic discovery using different curiosity styles. Inspired by human curiosity types—Busybody, Hunter, and Dancer—Rabbit Hole uses AI agents to simulate these distinct exploration strategies in academic research, generating summaries and novel insights. The platform aims to transform the way researchers navigate through literature by turning passive reading into an active, exploratory process that adapts to the researcher's curiosity style.

The Challenge

In academic research, the volume of information is overwhelming, making it difficult for researchers to explore new territories effectively. Conventional literature reviews can become time-consuming and often lack creative approaches to discovering unexpected connections. This problem is compounded by limited time and varying depth of focus, where researchers may need different levels of engagement with various topics.

Rabbit Hole was designed to address these challenges by:

1. Simulating Diverse Curiosity Styles: Enabling AI agents to explore research papers differently depending on curiosity style (broad vs. focused vs. creative).

2. Enhancing Literature Review Processes: Automating the process of diving into references, thereby saving time for researchers while presenting diverse viewpoints.

3. Providing Active Engagement: Making the research process more interactive by having agents mimic human curiosity, offering varied perspectives on the same academic content.

4. Helping Generate Novel Insights: Supporting researchers in developing creative, interdisciplinary connections that may not emerge through conventional means.

The Solution

Rabbit Hole combines LLMs and custom-built workflows to help researchers explore academic literature in a structured yet diverse manner:

Three Curiosity Styles

• Busybody: Broadly explores various references, forming loose and extensive knowledge networks, giving researchers a panoramic view of a topic.

• Hunter: Focuses intensely on specific subjects, maintaining strong coherence and forming tight knowledge networks, perfect for deep dives into specialized areas.

• Dancer: Takes creative leaps between disciplines, making unexpected connections to foster interdisciplinary understanding and insights.

AI-Driven Exploration

1. Research Paper Upload: Users upload a research paper (PDF) as the starting point.

1. Initial Analysis: AI agents analyze and summarize the document.

1. Reference Chaining: Each agent navigates references based on its curiosity style for five iterations.

1. Rationale Generation: Agents provide clear explanations for their paper selections.

1. Mega Summary: Each agent generates a comprehensive synthesis reflecting its unique approach.

1. Rating System: Users rate each summary's usefulness, helping identify the most effective curiosity style.

Here is a demo video of the whole process.

User Research & Analysis

• Conducted 20+ interviews with Stanford graduate students

• Implemented analytics to track preference patterns

• Gathered qualitative feedback on exploration styles

Key Findings

Quantitative Results

• Hunter style: Preferred by 71% of users

• Busybody style: Chosen by 29% of users

• Dancer style: Valued for innovative connections