DEMO / RECORDING
Overview & Result
Autonomous gaming agent playing Super Mario Bros in an emulator. Demonstrates Jev functioning as a high-frequency real-time reflex layer, sampling emulator screen coordinates and obstacle vectors every frame to make sub-50ms jump, run, and duck decisions.
How Jev fits in the loop
- Extract entity bounding boxes and platform terrain vectors from the emulator memory each frame.
- Feed immediate obstacle state into Jev Choice with candidate controller actions [JUMP, RUN_RIGHT, DUCK, WAIT].
- Receive top-choice action in 35ms with normalized action probability.
- Inject virtual controller button presses directly into the NES emulator loop.
How to reproduce
- Inspect author post and reproduction notes.
- Deploy agent loop and bind Jev decision endpoints.
- Run test inputs and observe latency and accuracy.
Why this build matters
Vivid proof that Jev can operate within continuous, real-time action loops where traditional 2-second LLMs would cause instant game over.
Limitations
- Community observed build; metrics are author-reported.
- Requires third-party dependencies as described in source.