Emilio Melis / Aatricks

C++ runtimes / JNI bridges / local inference

Edge AI. Native Performance.

I build edge AI software that runs locally on constrained devices, with emphasis on runtime ownership, memory control, and reliable native architectures.

Selected work GitHub profile

Selected work

Flagship system

Runtime target: Android-native multimodal inference

llmedge

Kotlin-owned orchestration over raw native engines, with practical control over memory, artifact lifecycle, and backend fallback.

Kotlin / JNI / C++ / Android NDK / Vulkan / OpenCL / ONNX

llmedge documentation homepage

Generation stack

Runtime shape: Local server, browser UI, and pipeline core

LightDiffusion-Next

A practical diffusion system shaped around local use, model flexibility, and implementation detail rather than benchmark theater.

Python / FastAPI / Streamlit / Gradio / PyTorch / Flux / SDXL

LightDiffusion-Next browser interface with prompt controls and generated output

Product surface

Product scope: Discovery, library, reader, and summary workflow

EasyReader

Productized Android UX backed by local model integration instead of remote summarization shortcuts.

Kotlin / Jetpack Compose / Room / Hilt / llmedge

EasyReader explore screen showing discovery cards on Android

More work

Core systems

CChess

C / engine logic / minimax / alpha-beta pruning / board evaluation