ProofLab AI

ProofLab AI — Turn the Real World into a Scientific Laboratory What if an ordinary smartphone video could become a reproducible scientific experiment? ProofLab AI is an open-source, multimodal research platform that transforms videos of physical phenomena into measurable evidence, falsifiable predictions, and scientific comparisons. Our flagship experiment studies a real physical pendulum through five stages: Prepare, Analyze, Understand, Predict, and Compare. Meta Muse Spark helps identify the experimental object and interpret scientific evidence. SAM 3.1 tracks the object's motion across video frames. A deterministic measurement engine calculates trajectories, angles, angular velocity, and oscillation periods using original video timestamps. The breakthrough is the complete scientific workflow: ProofLab analyzes an initial observation, proposes a follow-up experiment, freezes a numerical prediction before the second video's analysis, and compares the independently measured result against a predefined tolerance. Unlike conventional AI assistants that simply explain scientific concepts, ProofLab connects AI reasoning with physical observations and reproducible measurements. The platform preserves evidence, source hashes, uncertainty indicators, graphs, and scientific reports. Results can agree with a prediction, disagree, or remain inconclusive when the evidence is insufficient. Built in Peru during the Meta Global AI Developer Hackathon 2026, ProofLab AI demonstrates how multimodal AI can help make evidence-based experimentation more accessible. Our vision: a scientific laboratory powered by AI, accessible from an ordinary camera. Live platform: https://prooflab-ai.com/