Lucky Umoka

StepLint is an agentic physical-world procedure intelligence and verification system. StepLint turns an expert’s demonstration into a procedure a learner can follow and the system can check. AI can already explain how to perform a task, but reliably guiding and verifying someone as they physically perform it is still hard. Steplint turns an expert’s demonstration into executable, real-time safety rules. An expert records a task and talks through it (such as assembling hardware). StepLint combines the video, the voice transcript and object tracking to work out the starting state, the actions, the intermediate states and the intended result. Felicity, an AI agent, audits the draft procedure against that evidence and flags inconsistencies for the expert to review. Once the expert approves it, the procedure becomes a Practice experience. StepLint then supervises through a live webcam.During Practice, Gideon guides the learner while computer vision tracks the relevant objects. A deterministic verification engine evaluates whether each required action and the final goal have been achieved. If the user gets stuck, an asynchronous agent, Felicity, helps them. Everything was built solo during the hackathon; the repository's first commit is Oct 29, 2026. It is designed for procedures where doing the steps in the right order matters, such as equipment assembly, technical operations and maintenance training. The aim is to make that work easier to learn, more consistent to execute and easier to verify.