Announcing the winners of the first Arm AI Developer Challenge

The inaugural Arm AI Developer Challenge set out to explore what’s possible when AI applications are designed to run locally and efficiently on Arm-based mobile and edge devices. As AI workloads increasingly shift closer to users—driven by requirements for low latency, privacy, reliability, and energy efficiency—the challenge focused on practical, real-world approaches to on-device AI development.

In partnership with Devpost, the Arm AI Developer Challenge invited developers to build AI-powered applications optimized for Arm-based platforms, including smartphones, tablets, and single-board computers such as Raspberry Pi. Rather than relying on cloud-backed architectures, submissions emphasized offline or low-connectivity scenarios, where inference, data processing, and user interaction happen on the device.

Submitted projects demonstrated diverse approaches to on-device AI covering generative AI, computer vision, natural language processing, and geospatial analysis, while accounting for performance-per-watt, memory constraints, and usability.

“Developers are constantly looking for ways to challenge themselves,” said Prithvi Mattur, lead of Developer Growth at Arm and a judge for the challenge. “The Arm AI Developer challenge provided them with a new way to explore AI capabilities on device. It was all about empowering developers to go one level deeper and optimize diverse AI workloads across the devices we all use every day.”

Meet the winners

The challenge was evaluated by Arm engineers and developer evangelists with experience across software. Projects were assessed using four core criteria: technological implementation, user experience, potential impact, and the “wow” factor. After reviewing all submissions, the panel of judges selected six projects that demonstrated strong alignment with the challenge goals.

“One of the most impressive things was how creative the on-device AI solutions were,” said Gabriel Peterson, senior ML engineer and a judge of the challenge. “The submissions showed how Arm enables efficient, responsive mobile and edge apps while still giving developers room to experiment.”

Out of 142 submissions, the judging panel selected six winners of the Arm AI Developer Challenge, who together will receive $15,000 in total cash prizes for their outstanding innovation and execution on Arm-based hardware.

1^st prize

Chuck’it: An on-device bookmarking app that uses AI to automatically organize and semantically search anything you save—screenshots, links, text—so users can capture information instantly without breaking their flow or sacrificing privacy.

Han Wei Tan and Varun Bhalero, who both created Chuck’it, believe that receiving 1st prize in the Arm AI Dev challenge allows them to be more confident in the product they are trying to build. They said: “The next step is learning how to optimize on-device AI to squeeze more performance out of what already exists. These future endeavours will involve Arm-based devices in some way, with many systems we use on a day-to-day basis running on Arm.”

2^nd prize

DreamMeridian: GeoAI on Pi: A natural-language spatial query engine that runs on a low-power Arm-based Raspberry Pi 5, enabling humanitarian and field workers to find routes, distances, and nearby services from local map data without internet access, cloud compute, or GIS expertise.

Adam Munawar Rahman, who was a solo participant on the GeoAI on Pi project, spoke about the vital role of the Arm compute platform in powering reliable on-device AI that works without connectivity. He said: “Arm will remain central to my work: low cost, low power edge deployment is the only way to reach places where infrastructure fails. For humanitarian and emergency contexts, this changes what’s possible in the field.“

InstaMeme: A fully on-device, Arm64-optimized iOS app that uses local vision and language models to turn photos into shareable memes instantly, proving that fast, private, and creative AI generation can happen entirely on-device without cloud services.

Mark Foster, who developed the InstaMeme project, was delighted at how a tightly scoped AI model can perform on Arm-powered mobile devices. He said: “Performing AI inference on Arm-based mobile devices allows me to provide an offline experience for my app users. Local models offer several advantages over cloud-based approaches, including lower latency, reduce compute and infrastructure costs, improved privacy, and reliable offline experiences.”

3^rd prize

Jackqr: An on-device study tool that converts poorly scanned PDFs and textbook photos into clean, searchable learning materials with AI-powered simplification and flashcards, all running locally so students can study effectively without reliable connectivity. See the demo here.

Epictetus: A privacy-preserving Android chatbot leveraging AI through Arm KleidiAI, XNNPack and MediaPipe that offers practical guidance, emotional support and everyday advice to users while keeping conversations entirely on smartphone devices. See their demo here.

Pocket Garden: An AI-powered gardening companion that runs locally on Arm-based Raspberry Pis, using real-time environmental data and a multi-agent system to give new gardeners personalized, location-aware advice that helps them improve yields and stay engaged throughout the growing season. See the demo here.

All of the winning projects can be explored in more detail here.

Building the future of AI on Arm

Seeing this developer community turn Arm-powered platforms into practical AI experiences reinforces why Arm invests in programs like this: to remove friction, share the right tools and examples, and help great projects ship.

Across the winning submissions, several consistent themes emerged:

• Prioritizing privacy and trust by default, with developers choosing fully offline workflows that keep user data on device without cloud dependencies.
• Performance-per-watt as a core product feature, with developers using techniques such as model quantization, efficient runtimes, and right-sized models to deliver responsive user experiences on Arm-powered smartphones and Raspberry Pi systems.
• The ability to operate in real-world conditions where connectivity is unreliable or unavailable, with developers demonstrating how on-device AI can remain useful in classrooms, field operations, and other constrained environments.

Together, these projects illustrate how Arm architecture enables scalable, privacy-first, and resilient AI at the edge, helping accelerate adoption of on-device AI patterns that developers can build on and extend.  

What’s in store for the future with Arm and on-device AI?

The Arm AI Developer Challenge reinforces a clear message: AI capabilities are increasingly moving to on device, with the Arm compute platform at the center of this broad trend. As AI continues to evolve, developers will play a critical role in shaping applications that are not only powerful, but efficient, private, and accessible.

“Looking at these projects, it’s clear that we’re on the front line when it comes to optimizing AI to run on-device and on CPUs,” said Prithvi Mattur, a judge for the challenge. “Developers are considering Arm in their community as the preferred path for on-device AI innovation.”

Watch the Arm AI Developer Challenge Winning Showcase Webinar for more details about what made the winning entries stand out, how teams delivered strong implementation and UX and impact for AI running on Arm-powered devices, and how to set yourself up for success for future challenges.

To Devpost and everyone who participated, thank you for building, experimenting, and pushing boundaries. Arm will keep showing up with more challenges, content, and hands-on support, so developers can keep building what’s next on Arm.