Talks.

AI coding agents can feel like magic, but their core loop is straightforward: send a message to a model, parse and execute tool calls, return the results, and repeat. In this 90-minute deep dive, we build that engine from scratch in Java, live on stage, and point it at a real codebase. As the agent compiles successfully, passes tests, and still violates every convention the team agreed on, we iteratively add context, constraints, and feedback, mapping each step to the AI coding tools developers already use. Attendees leave with a practical mental model of how coding agents work, what can be layered around them for smoother and safer results, and where their honest limits remain.

Peer-reviewed research and findings from Anthropic, OpenAI, Google DeepMind, Apollo Research, and UC Berkeley show that frontier AI models exhibit emergent alignment failures across three categories: training-induced misalignment, in-context scheming, and multi-agent peer preservation. These are documented, reproduced, and published behaviors of shipped models, not theoretical risks. In the same eighteen months, mechanistic interpretability delivered a genuine breakthrough: researchers can now trace the circuits inside a frontier model as it reasons, watch it plan a rhyme before writing a line of poetry, and detect hidden goals it was trained to deny. Attendees leave with a clear map of the frontier: what current AI systems are already showing, what interpretability is beginning to reveal, and what these developments imply for the years ahead.

AI coding agents can feel like magic, but their core loop is straightforward: send a message to a model, parse and execute tool calls, return the results, and repeat. In this live Java deep dive, we build that engine from scratch and point it at a real codebase. As the agent compiles successfully, passes tests, and still violates the team's conventions, we iteratively add context, constraints, permissions, observability, safety guardrails, and feedback. Attendees leave with a practical mental model of how coding agents work, what can be layered around them for smoother and safer results, and where their limits remain.

AI coding agents can feel like magic, but their core loop is straightforward: send a message to a model, parse and execute tool calls, return the results, and repeat. In this hands-on workshop, we build that engine from scratch in Java and point it at a real codebase. As the agent compiles successfully, passes tests, and still violates the team's conventions, we iteratively add context, constraints, permissions, observability, safety guardrails, and feedback. Participants leave with a practical mental model of how coding agents work, what can be layered around them for smoother and safer results, and where their limits remain.

Designed for software engineers and developers with little or no ML experience, this interactive deep dive introduces foundational machine learning concepts through hands-on Python and scikit-learn demos. Participants learn the basics of supervised learning while building and evaluating a simple classification model, then explore how ML fits into real-world software development and production workflows. No prior machine learning experience is required; basic Python knowledge is recommended.

There are both exciting new possibilities for AI and serious worries about its impact on human safety as this technology advances at an astounding pace. Examples like AlphaFold show that AI can be useful, while high-profile mistakes in facial recognition and criminal justice show the risk of misaligned goals and harmful errors. This session dives into the fundamental problem of AI alignment: making sure increasingly autonomous systems strive for goals that align with human ideals. We explore the societal and technical aspects of AI alignment through real-world case studies, expert warnings, and potential future scenarios. Participants leave with a firm grasp on why just turning it off will not cut it, what urgency signals academics and governments are responding to, and how culture, policy, and technology can steer AI toward safer outcomes.

As we rapidly integrate artificial intelligence into every aspect of life, its transformative power carries profound ethical and practical risks. This talk explores AI alignment and safety through a critical, engaging lens. Attendees confront real-world case studies, from biased facial recognition systems to autonomous vehicle failures, to understand how misaligned AI can amplify inequalities, endanger lives, and distort democratic processes. The presentation closes with actionable best practices for IT professionals who want transparent, ethically grounded AI that serves human interests.

From smart fridges that cannot think to AI-generated soda flavors that taste more like marketing than innovation, this talk dives into AI washing: the trend of attaching "AI-powered" to products, startups, and pitch decks to ride the buzz. The session explores real and ridiculous examples of AI being used as a shortcut to credibility. For founders, investors, and developers, it is a prompt to laugh, think, and question the next AI-branded thing they see.

This deep-dive session is tailored for software engineers, developers, and tech enthusiasts who want to move beyond buzzwords and understand how machine learning integrates into real-world software development. Over two and a half hours, we unpack supervised learning, model training and evaluation, and the tools that power modern ML systems, including scikit-learn, MLflow, and GitHub Actions. The session gives a code-aware walkthrough of the ML lifecycle, from data preprocessing to model deployment, with enough context for ML-curious developers to collaborate with data teams and reason about production systems.

Curious about IoT but not sure where to start? I was in the same place. Armed with a Raspberry Pi Pico 2, no prior experience, and a useful amount of trial and error, I set out to build my first project: a burglar alarm. This beginner-friendly talk walks through the bumps, breakthroughs, and "why is this not working" moments of that first IoT build, from wiring sensors to getting the alarm working.

This two-hour deep dive is tailored for software engineers, developers, and tech enthusiasts who want to understand how machine learning integrates into real-world software development. We unpack supervised learning, model training and evaluation, and the tools and workflows behind modern ML systems, including scikit-learn, MLflow, and GitHub Actions. The session provides a comprehensive, code-aware walkthrough of the full ML lifecycle, from data preprocessing to model deployment, and shows how ML can fit responsibly into development workflows.