Duarte O. Carmo has published a Python implementation of Jev that uses only 25 lines of code. The script uses the llama-cpp-python library to run a GGUF model, such as Qwen3-0.6B-GGUF, locally.
Jev Implemented in 25 Lines of Python
The implementation works by passing a prompt to the model and extracting the logits from the final token. By converting these logits into probabilities across specific labels—such as "Legitimate," "Spam," and "Phishing"—the code determines the likelihood of different outcomes. This approach allows for local execution of the task.
The original post is a parody blog post.
Sources
- Jev in 25 Lines of Python （Hacker News Frontpage, 2026-09-23）