Technology & Core

Compile the application, not just the model.

MountAIn Core is the shared technical foundation beneath Ready, Onboard and IBEX. It brings the trained model, target constraints and surrounding application into one deployment path.

Check your deployment route

deploy_defect_vision.py

# pip install mountain-sdk
from mountain import Model, Pipeline, Target

model = Model.load("vision_defect_v3.tflite")

pipeline = (
    Pipeline(model)
    .preprocess(resize=224, normalize=True)
    .postprocess(decode_boxes(), nms=0.45)
    .compile(Target("alif_e7_dk"))  # SRAM-aware mapping
)

pipeline.benchmark()  # 2.1 MB peak SRAM · 34 fps @ 400 mW
pipeline.deploy()     # flashes the runtime to the board

Where MountAIn fits

Between the trained model and the deployed application.

A product has to coordinate the model with memory, sensors, processing, runtime behaviour, power limits and the target toolchain. MountAIn works inside that reality without asking teams to abandon the framework, silicon or product architecture already chosen.

The MountAIn Core stack

Four capabilities. One deployment path.

Workflow

Bring in a trained model, define the target and make deployment constraints visible early.

Compiler

Prepare and optimise the graph before the hardware backend compiler, across the complete application path.

Middleware

Connect model execution to data movement, processing stages and product behaviour.

MountAIn SDK

Tools and interfaces for evaluation, integration and validation, confirmed for each engagement.

Three routes to the edge

Designed around real constraints

Product conditions shape every decision.

Memory

Optimise for what the complete application can actually hold and move.

Power

Treat energy as a system constraint, not an afterthought.

Latency

Keep decisions close to the data when response time or connectivity matters.

Toolchain fit

Work with the target and its backend tools rather than forcing platform change.

Maintainability

Create a route engineering teams can test, understand and carry forward.

Good evidence

A number without conditions is not proof.

A useful result identifies the model, task, input, target, precision, toolchain and test conditions. It shows what changed, what remains in the product stack and what has not yet been proven.

Python and model-first teams begin with the model and success condition.
Embedded and product-first teams begin with the device stack and operating constraints.

Trained model

MountAIn Core

Deployed application

Start with one bounded problem.

Define one model, one target and one success condition.