AWS LA Summit 2025 - Won at AI League Workshop

Went to my first AWS LA Summit today. I expected talks and product demos. Instead, I ended up training and deploying an AI model in a workshop and walked out with $75 in credits as a winner!

The morning opened with a session on how Amazon Q, KiRo/MCP, and Bedrock agents fit into the software lifecycle. One-line takeaway: these aren’t side tools anymore; they’re becoming part of how teams actually build.

The most fun was the AI League workshop. The challenge was to build a simple LA tourism assistant: create a dataset, train in SageMaker, deploy, then evaluate. I finished 2nd on the AI leaderboard (tied score, placed by submission time) and won $75 in AWS credits. Good reminder that hands-on work sharpens understanding faster than any slide deck.

The afternoon went deep. HyperPod showed that training at scale is as much about efficiency and cost as accuracy. Hugging Face on Trainium made fine-tuning feel less about feasibility and more about optimisation.

Key takeaways:

• Bedrock is already in active use, not a “future vision.”
• Fine-tuning on SageMaker has moved from “can we do it?” to “how do we optimise cost, scale, and latency?”
• Training at scale is defined by cost/performance trade-offs as much as accuracy.

Grateful for the conversations and ideas shared today. Read more on LinkedIn →