Build On-Device AI

Name: Build On-Device AI
Rating: 4.5 (113 reviews)

Master On-Device AI! Learn to Train, Compile and Profile AI Models for Edge Device deployement with Qualcomm AI Hub

Created byStart-Tech Academy, Abhishek Bansal, Pukhraj Parikh

Last updated 4/2026

English

EnglishItalian [Auto],

What you'll learn

Understand the complete workflow of On-Device AI deployment, from training to inference
Learn how to use Qualcomm AI Hub for managing, compiling, and optimizing AI models
Master model profiling and compilation to enhance performance on edge devices
Learn quantization techniques to optimize AI models for mobile, IoT, and embedded systems
Understand the difference between symmetric and asymmetric quantization

Course content

7 sections • 24 lectures • 2h 1m total length

Introduction3:09
Explore on-device ai basics and practical optimization for resource-constrained devices, including training, compiling, profiling, and inference in Qualcomm ai hub, and learn quantization techniques for edge deployment.
Course Resources0:04

On-Device Introduction5:33
Discover how on-device AI runs AI models locally on smartphones and edge devices, delivering low latency, offline functionality, and stronger privacy, while facing resource and hardware diversity challenges.
Qualcomm AI Hub Introduction8:46
Discover how Qualcomm AI hub enables on-device AI on edge devices by converting and optimizing models for deployment on Qualcomm hardware with hardware accelerators for high performance and energy efficiency.
This is a milestone4:07
Celebrate reaching this milestone as you rank among the top 50% of learners and stay motivated to complete the course, using Q&A, AI assistant, and offline video downloads.
Qualcomm AI Hub Login2:32
Register for a Qualcomm AI hub API key, log in to obtain your API token, and deploy on-device models by bringing your own torch model or using Qualcomm's library.
Quiz

Steps for On-Device Deployment7:20
Learn how to prepare an AI model for on-device deployment through training, graph capture, compilation, validation, and performance profiling on edge hardware.
Model Training Phase - Theory8:02
Explore the training phase for on-device deployment with a pre-trained MobileNet v2, freeze parameters, and trace with TorchScript for a lean, reliable edge inference model.
Training the Model- Practical5:04
Install the Qualcomm AI hub libraries and pre-built models, configure the API token, verify devices, load MobileNet v2 for on-device inference, then freeze and trace the model to TorchScript.
Quiz

Compiling the Model - Theory8:09
Jump into the compilation step of on-device AI deployment, optimizing a Torchscript model with hardware-specific improvements, operator fusion, and quantization using the Qualcomm AI hub for TensorFlow Lite runtime.
Compiling the Model - Practical3:42
Compile the traced model for a target device using Qualcomm AI Hub, specifying the model, target device (Samsung Galaxy S24), and input specs, with TensorFlow Lite as the runtime.
Profiling the Model - Theory3:35
Profile the compiled model to optimize performance on the target device by measuring inference time, memory usage, compute unit utilization, and conducting layer-by-layer analysis.
Profiling the Model - Practical8:02
Explore how to run a profile job on Qualcomm AI hub for a compiled model, compare CPU, GPU, and NPU performance, and interpret timing and memory metrics.
About the upcoming role play.0:38
Interview Preparation: Explaining On-Device AI Deployment

Inference7:39
Perform on-device inference by preprocessing inputs and shaping to 1x3x224x224, then apply softmax for top-5 ImageNet predictions. Compare on-device results with cloud and torch models to validate performance.
Downloading the Model7:37
Download the compiled tflite MobileNet model for on-device deployment by code or the Qualcomm AI hub, and explore ready-to-use models in Qualcomm library and Hugging Face.

Introduction to Quantization6:12
Explore quantization to optimize on-device AI by converting 32-bit numbers to 8-bit integers, shrinking model size, speeding inference, and saving power, with notes on symmetric and asymmetric methods.
Symmetrics quantization9:14
Apply symmetric quantization to map fp32 values to a symmetric int8 range, compute the scaling factor, quantize and dequantize values, and assess quantization error in neural network matrices.
Asymmetrics Quantization10:16
Learn asymmetric quantization for on-device ai: compute a scale and zero point to map f32 values to int8, achieving full-range use and reduced quantization error versus symmetric quantization.
Quantization Techniques - Practical Application7:20
Learn how to apply on-device quantization with int8 weights and activations, using calibration data and Qualcomm AI hub's quantized models to deploy MobileNet v2 on devices.
New AI Features in On-Device AI (The Latest Updates You Must Know)1:25
Demonstrating Expertise in On-Device AI and Deployment Concepts during an Interview
The final milestone!1:33
Complete the build on-device ai course and join the top 5% of students. Unlock your certificate after the next lecture by marking all lectures complete and leaving a review.

Requirements

Basic Python knowledge is recommended, but no prior AI experience is required

Description

If you are a developer, data scientist, or AI enthusiast looking to create deployment-ready efficient AI models for edge devices, this course is for you. Do you want to accelerate AI inference while reducing computational overhead? Are you looking for practical techniques to optimize your models for mobile, IoT, and embedded systems?

This course will teach you how to train, compile, profile, and optimize AI models, ensuring they run efficiently on resource-constrained devices without compromising performance.

In this course, you will:

1. Learn the complete workflow of On-Device AI Deployment – from training to inference.
2. Understand Qualcomm AI Hub and how to use it for AI model management.
3. Explore model compilation and profiling to enhance performance.
4. Implement inference techniques for deploying models on edge devices.
5. Master quantization techniques to optimize AI models for low-power hardware.

Why Learn On-Device AI?

Deploying AI on edge devices allows you to reduce latency, enhance privacy, and optimize performance without depending on cloud computing. By mastering quantization, model profiling, and efficient AI deployment, you can ensure your models run faster, consume less power, and are ready for real-world applications like mobile AI, autonomous systems, and IoT.

Throughout the course, you'll gain hands-on experience with real-world AI deployment scenarios. You will balance theory and practical application to make your models leaner, smarter, and deployment-ready.

By the end of the course, you'll be equipped with the skills to train, optimize, and deploy AI models on edge devices, making you a valuable asset in the field of AI deployment.

Ready to take your AI models to the next level? Enroll now and start your journey!

Who this course is for:

Beginners in machine learning looking to gain hands-on experience in model optimization and on-device AI deployment
AI professionals, data scientists, and students who want to optimize models for deployment on resource-constrained devices like mobile, IoT, and embedded systems
Developers and engineers interested in learning how to use Qualcomm AI Hub to compile, profile, and deploy efficient AI models

Build On-Device AI

What you'll learn

Explore related topics

Course content

Introduction2 lectures • 3min

On-Device Introduction & Setup4 lectures • 21min

Model Training & Deployment Steps3 lectures • 20min

Model Compilation & Profiling6 lectures • 24min

Model Inference & Deployment2 lectures • 15min

Model Optimization & Quantization7 lectures • 36min

Conclusion2 lectures • 2min

Requirements

Description

Who this course is for: