Overview & result
Rizzo AI Academy shows rizzo-flow, built on Jev: The open, local take on Jev: typed decisions without generating a token.
From the original post
rizzo-flow by @rizzoaiacademy: The open, local take on Jev: typed decisions without generating a token.
How Jev fits in the loop
- Rizzo Flow downloads a fine-tuned 4B model and the llama.cpp runtime to your machine.
- You send unstructured state and typed questions to a local /v1/systemone endpoint.
- The model returns typed, probabilistic decisions without generating tokens.
- Everything runs locally.
How to reproduce
- Review the source documentation and repository at https://github.com/Rizzo-AI-Academy/rizzo-flow.
- Configure environment with TypeSafe Jev / System One API keys.
- Run test prompts against the decision endpoints.
Why this build matters
Shows Jev in practical use for research and data analysis.
Limitations
- On the typed-decisions test split the author measures 0.648 accuracy against 0.727 for TypeSafe Jev from its dataset card: "still below Jev in accuracy".
- Measured on the author's machine (RTX 5060 Ti).
- The default model download is 4.4 GB.