Welcome to the event

Welcome and event introduction

AI coding tools now let anyone - from non-coders to senior engineers - spin up an app in hours. But how do you go from “Tinder for pets” in a weekend to something real, stable, and safe? In this talk, Aashni Shah walks through the spectrum of AI development tools, live-builds a quick prototype, and shares lessons on when these tools are powerful, when they’re risky, and what it takes to turn an AI-generated prototype into a production-ready product. Key takeaways: \* How to choose the right AI coding tools for your skill level and project. \* Where rapid AI prototypes shine - and where they fail. \* What steps are needed to make an AI-built app secure, scalable, and maintainable.

Abstract - The increasing sophistication of Artificial Intelligence (AI) systems necessitates a rigorous, multi-dimensional evaluation paradigm that surpasses conventional automated metrics and subjective human assessments. This paper introduces Jo.E (Joint Evaluation), a structured evaluation framework that integrates human expertise, AI agents, and Large Language Models (LLMs) to systematically assess AI systems across critical dimensions: accuracy, robustness, fairness, and ethical compliance. Building on methodologies such as ”Agent-as-a-Judge” (Zhuge et al., 2024) and ”LLM-as-a-Judge” (Zheng et al., 2023), Jo.E provides a principled approach to identifying and mitigating AI risks through a tiered evaluation process. We validate this framework through controlled experiments on commercial models (GPT-4o, Llama 3.2, and Phi 3), demonstrating its capacity to detect model vulnerabilities that single-method evaluations miss. The framework’s key innovation lies in its structured information flow between evaluation tiers, enabling targeted human expert involvement where automated methods are insufficient. This creates a scalable, reproducible evaluation methodology with comprehensive coverage of critical AI safety dimensions. Our experimental results show that Jo.E successfully identified 22% more adversarial vulnerabilities and 18% more ethical concerns than standalone evaluation approaches while reducing human expert time requirements by 54%. Weblink - [https://himjoe.github.io/safealign/](https://himjoe.github.io/safealign/)

We will look at the many different ways that a model can be deployed to a Kubernetes environment. We'll compare NVIDIA NIM, Tensorizer streaming, in-cluster caching, and a standard container deployment from the perspective of effort, first-deploy performance, and scaling responsiveness. We'll finish with some recommendations and call outs for other factors to consider before deploying to production.

Closing remarks and next steps

Casual networking with food and drinks