Demo on PitchSpark. PitchSpark manages 5 data types including admin messages.
PitchSpark is the founder–investor matching platform built around an AI pitch scorer. Founders submit pitches and receive a calibrated 0–100 score across six dimensions (problem clarity, solution fit, market evidence, business model, team narrative, and communication). Investors discover scored pitches, request decks, and run the deal through a built-in pipeline. The platform also includes a community Roast Room for honest peer feedback and a public leaderboard ranking startups and investors by real platform engagement.
PitchSpark's AI pitch scorer uses a calibrated Claude Haiku 4.5 model with a versioned system prompt. Each pitch receives a composite score (0–100), letter grade, six per-dimension scores, and 80% confidence intervals. Weights are recalibrated weekly against real investor outcome data (deck requests, contact requests, funding events). A health-check job runs Wednesdays and auto-rolls back the active scoring config if drift is detected. The full rubric — with anchor descriptions per dimension — is at /llms-full.txt .