Kaleidoscope

Kaleidoscope is an AI-powered innovation intelligence engine designed to make AI-generated ideation measurable, diverse, and decision-ready. The project addresses a common problem with large language models: they can generate hundreds of ideas quickly, but many of those ideas repeat the same underlying concepts in different words. This creates a false sense of variety and makes it harder for teams to identify which ideas truly represent distinct opportunities.

The system reframes ideation as a selection problem, not just a generation problem. Instead of accepting every idea an AI model produces, Kaleidoscope breaks each candidate idea into six semantic dimensions: core mechanism, problem or opportunity, intended outcome, stakeholder, context, and value type. These dimensions allow the system to compare ideas based on meaning, not just wording.

Kaleidoscope uses embeddings and weighted similarity scoring to detect hidden overlap between ideas. Each new candidate is compared against the existing portfolio, and ideas that are too similar to previously accepted concepts are rejected. This process helps ensure the final portfolio contains ideas that are genuinely distinct, reducing redundancy while preserving novelty.

The project also includes a multi-agent workflow that moves ideas from generation to decision support. An Explorer Agent generates and screens concepts, a Refiner Agent expands the strongest ideas into structured concept briefs, and a Patent Validation Agent checks top ideas against patent data to surface early signs of intellectual property risk. These outputs give teams more than a list of ideas; they provide evidence, context, and prioritization signals.

The project was recognized as one of three finalist capstone projects selected to present at Carnegie Mellon University’s Tepper School of Business Data Nexus 2026, from a field of 30 competing projects. This recognition highlights Kaleidoscope’s contribution to improving the quality and breadth of AI-assisted innovation, showing how structured methods can turn raw idea generation into a more rigorous and useful decision-support process.