TTP-T: Accelerating Breakthrough Innovations: An End-to-End Platform for Large-Scale Discovery

This project is funded through the NSF Translation to Practice (TTP) program, which supports efforts to turn research discoveries into practical tools that benefit communities, industry, and society. Despite record spending on research and development (R&D), breakthrough discoveries are becoming rarer and harder to achieve. A key reason is that experts often get stuck thinking within their own field, missing solutions that already exist in other domains. History shows that some of the most transformative innovations come from borrowing ideas across fields. Today, these kinds of creative leaps happen mostly by luck, leaving organizations trapped in incremental improvements and rising R&D costs with diminishing returns. With this TTP project, an artificial intelligence (AI) platform will be developed that makes cross-domain discovery systematic and repeatable, giving R&D teams in industries such as automotive, consumer goods, and advanced manufacturing a way to generate more novel ideas, create stronger intellectual property, and move promising concepts to market faster. By increasing the effectiveness of existing R&D investments, the platform directly strengthens competitiveness of the United States in critical technology sectors. The platform will be tested and refined through structured pilots with corporate R&D partners, with the goal of launching a self-sustaining commercial enterprise. Beyond its commercial potential, the project trains the next generation of innovators by integrating the platform into university courses in design, business, and human-computer interaction, and it aims to make powerful innovation tools accessible to startups, nonprofits, and academic labs. This project creates an end-to-end, human–AI collaborative engine for solving complex research and development problems at scale by mining evidence-backed analogies from assorted domains. The system uses a novel agentic AI workflow to dynamically abstract a problem into its core structural components, searches millions of multimodal sources across the open web for analogous scenarios that address similar underlying challenges, and translates those discoveries into concrete, ranked "idea sparks" that are evaluated for novelty, impact, and feasibility. Unlike standard large language model prompting, which tends to produce repetitive and homogeneous ideas and rely on surface similarity, this approach optimizes for novelty and variety by finding and transferring underlying structural mechanisms across multiple domains. A key technical contribution is bridging the gap from raw inspiration to actionable concepts through automated de-risking, where multi-agent workflows synthesize evidence from patents, scientific literature, and market data to evaluate feasibility and surface trade-offs. The project also advances human-AI interaction by developing interactive tools that enable domain experts to navigate thousands of candidate solutions and steer the AI's analysis. The platform will be validated through structured pilots with industry partners, user studies with professional research and development professionals, and innovation challenges, measuring impact via expert-rated novelty, value, and feasibility of generated concepts alongside usability and creativity support metrics. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2553034 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Nikolas Martelaro | Institution: Carnegie Mellon University, PITTSBURGH, PA | Award Amount: $1,200,000 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2553034 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2553034.html