AMD AI DevMaster Hackathon 2026 Rules | Generated by AI

Home 2026.07

Question: What is this AMD AI DevMaster Hackathon document?

Answer:

This is the rule/specification document for the AMD AI DevMaster Hackathon 2026. It is a competition sponsored by AMD where developers build AI applications optimized for AMD Radeon GPUs + ROCm.

The core idea:

Build something useful with AI, but make sure the heavy compute runs locally on AMD GPUs using ROCm.

There are 3 tracks:

Track 1 — Multimodal AI Content Creation (probably most relevant for you)

Build AI tools around:

image generation
image editing
style transfer
video stylization
text-to-video
image-to-video
enhancement / upscaling
batch content creation

Examples:

Input:
  photo/video/text prompt

GPU pipeline:
  Stable Diffusion / FLUX / ControlNet / LoRA
  + ROCm acceleration

Output:
  anime style video
  cyberpunk video
  marketing content
  short video generation

Required stack:

AMD Radeon GPU from Radeon cloud
ROCm
PyTorch ROCm
ComfyUI / Diffusers / Transformers allowed
OpenCV / FFmpeg / Pillow allowed

Closed APIs cannot be the core. At least one key inference path must run locally on AMD GPU.

Deliverables:

source code
README
PDF project description
3–5 min demo video

Track 2 — Private Local AI Agent (also very suitable for you)

Basically:

Build a local ChatGPT-like agent running on AMD GPU.

Examples:

personal assistant
coding agent
knowledge base agent
document/email assistant
life management agent

Required capabilities:

At least 2 of these 5:

RAG
Tool calling
Multi-step planning
Local memory
Permission/privacy control

Possible architecture:

                User
                 |
             Web UI / CLI
                 |
             Agent Router
                 |
       +---------+---------+
       |                   |
    Local LLM          Tools
   Qwen/Llama          |
       |               |
   ROCm/vLLM       Browser
                   Files
                   Git
                   Calendar

        AMD Radeon GPU

Recommended:

vLLM
llama.cpp
Transformers + PyTorch ROCm
LangChain/LlamaIndex/Dify/etc.

Given your existing ww, iclaw, agent experiments, this track maps very closely.

Track 3 — Physical AI / Robotics

Robotics simulation:

robot arms
humanoids
quadrupeds
autonomous driving

Using:

Genesis
MuJoCo
LeRobot
OpenVLA

Need simulation + AI control + AMD GPU acceleration.

Probably less aligned unless you want to enter robotics.

Prize

Each track:

Place	Prize
1st	$5,000
2nd	$3,500
3rd	$1,500

Timeline

Important dates:

Registration opened: July 10, 2026
Submission opened: July 15, 2026
Deadline: August 6, 2026 23:59 Beijing/Singapore time

My recommendation for you

Given your background:

GPT-2 training
nanoGPT
agent CLI tools
blog/video automation
AMD ROCm interest

I would choose Track 2.

A strong project idea:

“OpenClaw Local AI Agent”

AMD Radeon GPU
        |
     Qwen3 / Llama
        |
   Agent Runtime
        |
 +------+-------+
 |      |       |
RAG   Tools   Memory
 |      |       |
Blog  Git    Vector DB
PDF   Shell  SQLite
Video

Demo:

User asks:

“Create today’s AI research video”
Agent:

search papers
      |
summarize
      |
generate markdown
      |
generate image
      |
generate video
      |
publish

Everything local on AMD GPU.

This directly demonstrates:

local inference
tool calling
workflow orchestration
practical value

The judges explicitly score these capabilities.

For winning, don’t build another chatbot. Build a personal AI operating system running on Radeon.

Back Donate