jevlike-esp32
Experiment running a trained jevlike option scorer as ESP-IDF firmware on an ESP32 microcontroller.
Overview
jevlike-esp32 is a small set of experiments porting the jevlike one-pass option scorer to embedded hardware. Training stays in Python using the jevlike repository; the trained scorer can then run either in Python or as ESP-IDF firmware on an ESP32, where it acts as a simple edge router choosing among text options.
The list summary describes a C scorer and a host-side check that compares firmware output with the Python reference. It is aimed at makers and researchers curious whether Jev-style typed decisions can run on a microcontroller without a network call. The README is brief and points to an esp32_demo folder for details.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which route or option best matches this input text (edge router)?(inferred) | — |
How it works
- Train a jevlike text scorer in Python
- Export the weights for the C scorer
- Flash ESP-IDF firmware to the ESP32
- Compare on-device outputs with the host-side Python scorer
Strengths
- Decisions run on-device with no network
- Host-side check verifies the port matches Python
Caveats
- Experimental demo with minimal documentation
- Model capacity is limited by microcontroller memory
- No accuracy figures given
Getting started
See esp32_demo/README.md in the repository; training uses the vinnylarouge/jevlike repository.At a glance
- Type
- Project
- Status
- prototype
- Language
- C
- Runtime
- ESP-IDF firmware on ESP32
- Has code
- yes
Use cases and integrations
How curated lists describe it
Edge deployment: exports a jevlike scorer as ESP32 firmware with a C scorer and a host-side check, putting one-pass decisions on a microcontroller.
yibie/awesome-jev
Frequently asked questions
What runs on the ESP32?
A C implementation of a trained jevlike scorer that picks among text options.
Does it call the Jev API?
No, it runs a small local model on the device.
Where does training happen?
In Python, using the original jevlike repository.
Related projects
Mangaba AI's local, calibrated typed-decision model for Portuguese text, trained to optimize the Brier score; proprietary since v5.
ProjectproductionPython- blink★ 22
Open one-pass typed-decision model with a dependency-free C99 runtime that also builds to a 66 KB WebAssembly module.
ServicebetaC - Foq★ 11
Local open alternative to Jev using a 1.58-bit ternary 8B model to answer booleans, choices, scores and Pydantic extraction in ~25 ms.
ProjectbetaPython Runs the open Laya decision model entirely in a browser tab with WebGPU and ONNX Runtime Web: no server, no API key.
ServicebetaTypeScriptllama.cpp-based C++ server that turns any LLM into a System One decision model, reading answers from next-token logits.
ProjectbetaC++Three open 4B decision models for English, Turkish and German that serve the Jev /v1/systemone API on a CPU in about 3 GB of RAM.
ServicebetaPython