TempoDance AI

TempoDance AI is the dance coach that learns how you learn.

"Dance tutorials can show the move, but they cannot see why I keep missing it. Rewatching at half speed gives every learner the same instruction, even when our errors are different."

"TempoDance watches body geometry, not body size. It compares normalized limb directions with the built-in reference, finds the limb that costs the most score, and selects a focused coaching rule for that error."

1. Leave Demo mode selected and press the circular play button.
2. Point out the reference and learner skeletons plus the per-bone error colors.
3. Let one loop finish; show the agent memory naming the repeated weak motion.
4. Show policy version increment and the cue changing for the next loop.
5. Let mastery stabilize; show the reasoned speed increase from 0.5x to 0.6x.

"This is not a chat wrapper. After each completed server-side loop, the agent compares the targeted bone's median with its prior baseline and initializes, retains, revises, or holds a predefined focus and coaching-rule strategy. A reliable improvement of at least 1.5 percentage points retains the intervention; scoring weights and mastery gates stay fixed."

"In this localhost demo, camera frames go to the local FastAPI process for in-memory inference and are not persisted by application code. Missing or low-confidence bones reduce score and coverage; loops below the coverage or consistency gates cannot automatically promote speed. The scripted demo removes camera, pose-model-download, and external-provider risk."

"Today it learns how to teach one eight-count. Next it becomes a reusable motor-learning layer for more dance styles and sport practice. TempoDance is the coach that learns how you learn."

Built With

• local-llm
• yolo
• yt-dlp