Introducing CRACQ

Granter.ai was my first internship. I joined as an AI R&D intern, Granter helps companies apply for grants more easily using AI agents. My job was to explore whether there was a research gap in how AI-generated documents were being evaluated. There was. So over the course of three and a half months, I researched the landscape and developed CRACQ, a multi-dimensional evaluation framework that scores documents across five traits: Coherence, Rigor, Appropriateness, Completeness, and Quality. Unlike single-score approaches, CRACQ combines linguistic, semantic, and structural signals to give both a holistic score and trait-level breakdowns. I trained it on 500 synthetic grant proposals and benchmarked it against an LLM-as-a-judge. Results showed CRACQ produces more stable and interpretable evaluations than direct LLM scoring.