Tag Archives: yow

Day 1 – Thu, Dec 12th 2024

Nilesh Makwana – Bridging the Digital Divide: A Path to Inclusive Innovation and Prosperity

Nilesh threw down the gauntlet, daring the room to confront their assumptions about inclusivity. He challenged them to reimagine their teams, their client targets, and their customer bases in a way that pushed against the very grain of the room’s prevailing mindset. Drawing on his own deep commitment to inclusivity, he shared powerful examples from his work with First Nations Communities, Local Councils, and people with disabilities, showcasing how he’d woven these principles into the very fabric of his teams.

Rachel Laycock – A New Era for Legacy Modernisation

This talk explored the challenges of legacy code modernisation and the potential of generative AI (Gen AI) to revolutionise the process. Legacy systems are complex, poorly documented, and resistant to change. Gen AI offers a new approach that can understand unstructured data, reverse engineer code, and generate documentation. This can empower developers to explore and modify legacy systems much faster and with fewer errors. The potential benefits include faster bug fixing, increased developer productivity, and a deeper understanding of complex codebases.

Thomas Vitale – Concerto for Java and AI – Building Production-Ready LLM Applications

This talk dived into building real-world applications with generative AI. It flipped the script on traditional development, starting with user experience. Imagine a composer’s assistant that understands emotions and suggests music based on movie scenes, director notes, and even free-form transcripts. This talk used real code examples to show how to build AI pipelines that spark creativity, not replace it. The audience learnt about vector search, semantic understanding, and how to leverage AI as a tool to empower human artistry. Added to that was Thomas’ live music performance that is weaved into the final output.

Marty Pitt – Adaptive Architectures – Building API Layers that Build Themselves

Tired of brittle, point-to-point integrations that break with every producer update? Taxi and Orbital offer a revolutionary approach, moving beyond traditional glue code and endless mapping. By focusing on semantics—the meaning of data—rather than just structure, Taxi provides a powerful language to define relationships between disparate systems. Combined with Orbital, an open-source UI, this dynamic duo empowers developers to connect data sources on demand, adapt seamlessly to change, and query information by its meaning, not just field names. Imagine a world where integration logic is generated automatically, where schema changes are handled gracefully, and where you can ask for exactly the data you need, regardless of its underlying structure. This is the promise of Taxi and Orbital: a world where integration is no longer a bottleneck, but a powerful enabler of innovation.

Joakim Sundén – How Autonomy Saved One of Spotify’s Most Loved Features From Being Killed

Joakim Sunden’s story reveals how Spotify’s autonomous squads saved its beloved Discovery Weekly feature from the chopping block. Faced with the challenge of helping users discover new music beyond simple search, Spotify experimented with various approaches, from human curation to complex algorithms. A small team, empowered by Spotify’s unique structure, resurrected a discarded “Play it Forward” hackathon project, transforming it into the personalised mixtape we know today. This wasn’t a top-down mandate, but a bottom-up revolution fuelled by trust, rapid experimentation, and a culture that embraced failure as a learning opportunity. Despite initial skepticism from leadership and scaling hurdles, Discover Weekly exploded in popularity, proving that true innovation thrives in environments where teams are given the freedom to explore, experiment, and ultimately, surprise the world.

Steve Smith – Understanding Rust; or how I learned to stop worrying and love the borrow-checker

Steve Smith’s talk dives into the world of Rust, a systems programming language that tackles memory safety and concurrency head-on. He demystifies Rust’s infamous “borrow checker,” explaining how it prevents race conditions by enforcing strict ownership and borrowing rules. From managing resources with Drop and exploring stack vs. heap allocation with Box, to navigating the complexities of lifetimes, Rc, Arc, and the Send/Sync traits for thread safety, Smith illuminates Rust’s powerful features. He even touches on the current state of asynchronous programming in Rust, advising caution for production use. This talk was an engaging tour through Rust’s unique approach to building robust and performant software.

Allen Holub – Getting Buy-In: Overcoming Larman’s Law

In an illustration of organisational transformation, a group of engineers defied the odds and overcame the stifling status quo. Recognising the limitations of traditional methods, they championed experimentation and a culture of “ask for forgiveness, not permission.” By focusing on clear business cases and demonstrably reducing risk through techniques like mobbing (multiple programmers working together) and queue management, they chipped away at ingrained resistance. This bottom-up approach, driven by a passion for efficiency and a willingness to learn from mistakes, ultimately led to a more productive and collaborative work environment.

Day 2 – Fri, Dec 13th 2024

Holly Cummins – The Efficiency Paradox and How to Save Yourself and the World

Looking at the technical history of the steam engines of the 1700s to modern software, the pursuit of minimising waste has been a central theme. Holly’s talk explored how everything from pre-cloud provisioning nightmares to e-waste and slow code contributed to a global inefficiency crisis. Solutions emerged, from LightSwitchOps for server management to optimising code and embracing build-time initialisation. Looking at the Vroom Model, Holly declared that faster code saves the world, linking execution time directly to energy consumption. But the narrative took a turn, questioning the very definition of efficiency. Jevons Paradox and queuing theory revealed that sometimes, slack—in systems and in human lives—is not waste, but a crucial component for resilience, productivity, and even happiness. The talk concluded with a powerful message: by examining our assumptions, working less, and embracing idleness, we can paradoxically achieve more.

Holly Cummins – Tradeoffs, Bad Science, and Polar Bears – The World of Java Optimisation

In a fascinating talk about performance optimisation, we explored the past pitfalls and paradoxes that have plagued this field. We learned that blindly chasing speed can backfire, as Akamai discovered with a 100ms delay leading to a 7% drop in conversation rates. Even high-frequency trading firms lose millions due to platform delays. The key takeaway? We must optimise for the right thing, at the right time. Throughput, response time, latency, and resource utilisation are all crucial aspects to consider. Interestingly, Google once used station wagons to deliver data tapes for maximum throughput, highlighting the trade-offs involved. Furthermore, requirements change. An API optimised for desktop use might crumble under the load of mobile users. The talk also emphasised the importance of measurement over intuition. We saw how the McNamara Fallacy can mislead us to focus on lagging indicators that are easy to measure but hard to improve, instead of leading indicators that might be harder to grasp but offer more control. The concept of “micro-optimisation theatre” was another fascinating point. Spending hours tweaking code that has minimal impact is a waste of time. Finally, the talk addressed the allure of the “shiny stuff” like virtual threads, which may not be helpful for all situations. The core message? Optimise with data-driven insights, not fads.

Lu Wilson – Beyond Chat: Bringing Models to The Canvas

In a rapid-pace tour of the past and future of human-computer interaction, Lu talked about how their team has explored how LLM and generative AI interfaces might move beyond text. Lu started by revisiting the limitations of early computer interactions and the paradigm shift brought about by the Macintosh’s graphical interface. Next, they highlighted the text-based limitations of current AI tools like ChatGPT and proposed a “canvas” approach where users and AI collaborate on a visual workspace. This canvas could be used for many purposes, from generating websites with drawings to creating flowcharts and data visualisations. Lu’s team of researchers even experimented with using a programming language under the hood to instruct the AI and achieved some impressive results. They believe this canvas approach holds immense potential for the future, allowing us to interact with AI in richer and more intuitive ways.

Rod Johnson – Practical Gen AI: Building a Chatbot using Spring AI

In this talk Rod explored how Spring, a mature framework, is being adapted for the emerging world of AI. Spring is well adapted to this new challenge due to its structure and principles. A strong community and core team who’ve kept the framework relevant for over 20 years. Spring’s core principles of dependency injection, portable service abstractions, and aspect-oriented programming (AOP) are proving valuable for building AI applications. Rod showed how Spring integrates with tools like OpenAI and Ollama, allowing developers to leverage powerful AI models. While large language models (LLMs) like GPT-4 offer impressive capabilities, their black box nature creates challenges. The solution in this talk, was that Spring’s framework helps break down interactions into smaller, more manageable pieces. Rod also explored how to integrate factual information retrieval using vector databases like Neo4J to ground chatbots in reality. Finally, the talk showcased how Spring’s features like topic guards and advisor chains can be used to create robust and informative chat experiences. This approach demonstrates the continued relevance of Spring in the ever-evolving world of AI.

Rasmus Lystrøm – How to Lead your Organisation’s AI-transformation: Strategies, Skills, and Culture

Rasmus deconstructed the hype surrounding enterprise AI adoption, arguing that many organisations fall into predictable traps. He then debunked common strategies like building internal “ChatGPT” replicas or creating isolated AI Centers of Excellence, highlighting their inefficiency and tendency to create more problems than they solve. The core message emphasised that simply buying AI tools doesn’t equal transformation; true value comes from organisational change. Rasmus advocated for empowering full-stack teams, granting them ownership and autonomy, and focusing on business outcomes rather than chasing the latest tech. He then criticised the tendency to optimise prematurely and the dilution of solutions by traditional IT processes. Ultimately, Rasmus urged a shift towards building small, functional AI solutions, getting them into production quickly, and iteratively expanding them—a path that prioritises practical application over grand, often failing, centralised projects. Rasmus concluded by questioning the current direction of AI development, suggesting it should focus more on automating menial tasks rather than focusing on creative outputs, which are more interesting to AI developers.

Ben Sadeghipour – AI Powered Bug Hunting

Ben, a white-hat researcher, driven by insatiable curiosity, has amassed a $1M bounty haul since 2022, discovering vulnerabilities for giants like Airbnb, Zoom, and Apple. Leveraging the power of AI, he effortlessly navigated the digital landscape, bypassing traditional research methods like Stack Overflow to use LLMs like Claude and ChatGPT to generate malicious payloads and categorise vast troves of data, such as NASA’s 20,000 subdomains. From uncovering exposed GitLab repositories and exploiting insecure IIS Direct Object References to employing techniques like Zip Slip and SSRF attacks (even accessing AWS metadata and exploring PDF renderers at major companies), Ben demonstrated the immense potential of AI in both offense and defense, effectively using it as a “second brain” to uncover and exploit weaknesses that would have been impossible to find alone.