Scaling Edge AI deployment for face recognition across embedded and mobile platforms

How we helped [Client] achieve [Result] with our [Technology/Service]

Anonymised client

Project Overview

Client type

global technology provider

Industry

biometrics identity tech

Hardware platforms

Ambarella CVflow Rockchip Hailo Qualcomm iOS Android

Project type

managed team

Device type

embedded biometric devices phones and tablets

Engagement model

extension of the customer’s in-house team

Timeline

4+ years ongoing

Location

Bay Area US

Client

The client is a global leader in face-recognition technology. Estigiti supported the embedded and mobile side of that portfolio, especially where on-device CV/ML pipelines had to hold up across more than one runtime and hardware environment.

Objectives

The work covered a broad delivery surface: software development, DevOps, verification QA, technical product management, and data collection support. The operational goal was to keep the delivery backbone credible across multiple embedded and mobile targets over time.

Challenges

Keeping biometric pipelines consistent across many platforms, handling structured-light anti-spoofing, validating performance in hardware-in-the-loop, and supporting downstream customer integrations without fragmenting the engineering effort.

Solution

Estigiti operated as a managed team embedded into the client’s broader product effort. That team covered embedded software, mobile delivery, CI/CD, QA, data-support tooling, and technical product leadership. On the device side, the work ran across Ambarella CVflow, Rockchip RKNN, Hailo, Qualcomm QNN, iOS CoreML, and Android TFLite.

The service also included bespoke embedded software, structured IR-light processing, test infrastructure, and cloud/backend integration.

Technologies

Embedded vision platforms

Ambarella CVflow: CV22, CV25, CV28 Qualcomm QNN Rockchip RKNN: RV11xx, RK3588 Hailo: H8, H15

Mobile AI

iOS CoreML Android TFLite phone and tablet targets

Biometric pipeline

face recognition anti-spoofing structured IR light processing

Delivery backbone

C++, Python, Swift, Kotlin, PyTorch CircleCI hardware-in-the-loop performance testing pytest and googletest REST APIs, K8S, GCP

Results & Benefits

Broader target-platform coverage

One programme spanning embedded boards and mobile endpoints.

Stronger release discipline

CI/CD and hardware-in-the-loop testing helped keep deployment quality visible.

Better anti-spoofing path

The biometric scope covered more than recognition alone.

Longer-term delivery continuity

The team supported the programme over more than four years and multiple project needs.

Other Case Studies

???

Guiding an access-biometrics device from platform choice to edge integration

Machine vision consulting and embedded vision advisory for a team bringing face biometrics into ticketing, access control, frictionless payment, and credentialing on real edge devices.

Read more

???

Integrating face biometrics into a custom Ambarella access-control platform

Machine vision integration and embedded vision engineering for a next-generation access system, from video quality and matching performance to wired and wireless control-system links.

Read more

Discuss Edge AI deployment

If your model works in isolation but not yet across the real device landscape, we can help with the runtime, pipeline, and delivery layers that make deployment stick.