🧠

eAI

AI/ML Inference Engine

On-device AI/ML inference engine enabling machine learning without cloud connectivity. Supports neural networks for classification, object detection, NLP, and more.

Intelligence LayerC/C++, PythonActive Development

View on GitHub

Key Features

Tiny Runtime — Under 64 KB ROM

TFLite Micro, ONNX Micro, custom eAI format

Hardware Acceleration — CMSIS-NN, NEON, RISC-V vector

Model Zoo — Keyword, anomaly, image, health, gesture

On-Device Training — Federated & transfer learning

INT8/FP16 Quantization

Python Training Pipeline — PyTorch & TensorFlow

Architecture

Python Training Pipeline (PyTorch/TensorFlow)
├── eAI Model Compiler (Graph Optimization, Operator Fusion)
├── eAI Inference Runtime (C)
│   ├── Model Loader
│   ├── Operator Registry
│   └── Memory Manager
├── Hardware Abstraction (CMSIS-NN, NEON, RV Vector)
└── eos Integration (Task, Timer, DMA)

Code Example

#include <eai/eai.h>
#include <eai/model.h>

eai_runtime_init();
eai_model_t *model = eai_model_load(
    "/flash/models/hr_quality.eai");

float ppg_data[128];
read_ppg_sensor(ppg_data, 128);

eai_tensor_t input = {
    .data = ppg_data,
    .shape = {1, 128},
    .dtype = EAI_FLOAT32
};

eai_tensor_t output;
eai_infer(model, &input, &output);
float quality = ((float *)output.data)[0];

API Highlights

Function	Description
`eai_runtime_init()`	Initialize inference runtime
`eai_model_load()`	Load model from flash/file
`eai_infer()`	Run inference on input tensor
`eai_profile()`	Profile inference performance
`eai_quantize()`	Dynamic INT8 quantization

← Back to All Products