​Why Attend?

​​Join the world’s leading AI engineers, researchers, and enthusiasts in the heart of Paris for a 2-day intensive hackathon and workshop, focused on optimizing LLM inference performance using AMD’s powerful MI300X GPUs. Whether you're a systems engineer, an ML researcher, an AI enthusiast or a student this is your chance to:

• ​​Get hands-on with vLLM, SGLang & kernel optimization

• ​​Showcase your skills in LLM pipeline tuning

• ​​Compete for performance-based prizes

• ​​Learn from AMD, Hugging Face, Meta & more! 

​​Seats are limited. We encourage you to register in advance to secure your spot.

​Agenda:

​​Day 1:

• ​​9:30 AM – Doors Open

• ​​10:00 AM – Opening Talk by Anush on "What’s New with AMD’s Developer Initiatives?"

• ​​10:15 AM – Hackathon/ Competition Introduction

• ​​10:30 AM – Workshop 1: Optimizing Model Serving with vLLM

• ​​11:15 AM - 11:30 AM Break– Workshop 2: Developing Optimized Kernels

• ​​12:15 PM – Lunch + Networking + Hackathon/ Competition Team Formation if not done already

• ​​1:00 PM – Hackathon Kickoff+ Overview on Rules, Resources & Support

• ​​6:00 PM – Dinner & Networking

​​Day 2

• ​​8:00 AM – Doors Open + Breakfast Networking

• ​​8:45 AM – Partner Talk: Hugging Face- {Details Coming soon}

• ​​9:00 AM – 2:00 PM – Hackathon Continues

• ​​12:00 PM – 1:00 PM – Lunch

• ​​2:00 PM – 4:00 PM - Evaluation

• ​​4:00 PM – 5:00 PM – Optional Top 10 Team Presentations (5 min each)

• ​​5:30 PM – 8:00 PM – Awards + Happy Hour

​Workshop Details:

​Building AI Agents on AMD GPUs

​This hands-on workshop teaches you how to set up AI models and build intelligent agents (like chatbots or task-automation tools) using AMD MI300X GPUs. You’ll learn from AMD experts on how open-source tools like Qwen3, Llama4, and a browser-based workflow. No prior experience needed!

​ Fine-Tuning AI Models

​This second beginner-level workshop focuses on fine-tuning pre-trained AI models to make them smarter or more specific to a task (like better answering questions or generating images). You’ll use AMD GPUs, torchtune, and other open-source tools like Unsloth, with guidance from AMD engineers.

​Hackathon Details

• ​1st Place: ₹3,00,000 per team

• ​2nd Place: ₹2,00,000 per team

• ​3rd Place: ₹1,00,000 per team

​Track 1: AMD AI Premier League (AAIPL)

​A head-to-head AI challenge where teams of up to 3 developers build two language model based agents:

• ​Q-agent: Generates valid, challenging multiple choice questions on puzzles and mental ability.

• ​A-agent: Attempts to answer questions posed by the opposing team’s Q-agent.

​Goal: Create the most difficult yet solvable questions while ensuring your A-agent can accurately answer them.

​Format:

• ​Matches are played between teams where Q-agents pose questions and A-agents respond.

• ​Winning teams advance to the next stage.

​Resources Provided:

• ​1 MI300 GPU for 24 hours

• ​Sample code for prompt tuning, reinforcement learning, and fine-tuning

​Deliverables:

• ​Q-agent.py: Outputs a JSON file of generated questions

• ​A-agent.py: Outputs answers

• ​PowerPoint summarizing techniques used

​Track 2: Agentic AI Scheduling Assistant

​Problem: Scheduling meetings across time zones is inefficient and requires multiple back-and-forth messages. Existing tools lack intelligent conflict resolution.

​Goal: Build an AI-powered scheduling assistant using Agentic AI that automates meeting coordination by

• ​Scanning participant calendars to find optimal meeting slots

• ​Negotiating conflicts with other participant agents

• ​Sending polite reschedule requests

• ​Instantly confirming meetings or proposing alternative times

​Expectation: A seamless, intelligent assistant that minimizes manual effort in scheduling.