Prompt Engineering Conference

Keynote: Beyond Output: Evaluating Agentic Workflows in LLM Systems

As LLMs evolve into multi-step agents, evaluation must move beyond input-output correctness. This session shows how to assess both node-level decisions and full agentic flows, with a live demo of multi-node agents and practical eval techniques.

A Comprehensive Guide to Prompt Engineering Speaker participation

Prompt engineering is rapidly emerging as a critical skill for anyone working with Large Language Models. This session will demystify the process, blending the art of intuitive communication with the science of structured querying. We will cover core concepts such as prompt structure, context injection, and task specification, alongside practical strategies for eliciting creative text generation, accurate information retrieval, and effective problem-solving from LLMs. Learn how to master the nuances of prompting and unlock the full potential of these powerful AI tools.

Dr. Roushanak Rahmat is a Google Developer Expert and AI Strategy Leader with a PhD in Artificial Intelligence, specialising in computer vision and image processing. She has designed and delivered AI and Generative AI solutions for leading organisations including IBM, NHS, Elekta, Coca-Cola, Lloyds, and Scottish Power — turning complex challenges into impactful business outcomes. Currently serving as Lead Data Scientist at IBM Consulting, Dr. Rahmat drives AI and GenAI strategies for clients in healthcare, finance, and energy. Previously at Elekta, she spearheaded AI innovation projects to enhance radiotherapy treatment through advanced deep learning models and multimodal imaging data. Her technical expertise spans MRI, CT, CBCT, and large-scale data analysis using Python, PyTorch, Keras, TensorFlow, and Jax. Beyond her technical leadership, she is a passionate advocate for AI ethics, diversity, and inclusive innovation. As a Women Techmakers Ambassador and an active public speaker, she shares insights on AI development, adoption, and strategy through Medium articles, YouTube videos, and open-source contributions. Dr. Rahmat excels at bridging the gap between technical teams and strategic stakeholders, fostering collaboration, and making AI accessible to all. Portfolio: roushanakrahmat.github.io

Build without limits - Tips and techniques for better Prompting

AI Coding Assistants promise to revolutionise developer productivity, yet many engineers struggle to achieve consistent results. Drawing from my twelve month journey with these tools, this talk reveals practical strategies for transforming frustrating experiences into reliable outcomes. I'll share specific techniques that have consistently produced good quality code, concrete examples of successful prompting patterns, and resources that developers can use. Whether you're new to AI assistants or looking to refine your approach, you'll leave with a better understanding of the potential of these tools' to improve your development efficiency.

I am a developer advocate for open source, currently working at AWS. I am passionate about, and help customers be successful in using open source.

Beyond Handcrafted Prompts: The Era of Prompt Optimization

Prompts are powerful, but they are also brittle. Manually handcrafted prompts often work well for a time, but when models or their weights change, performance degrades forcing developers back into a cycle of prompt tinkering. This trial-and-error process is laborious, unscalable, and costly to maintain. In this talk, we introduce the era of prompt optimization: what it is, how it works, and why it is essential for building scalable and reliable AI systems. We will explore current tools and techniques, with a focus on automated prompt optimization using DSPy and its family of optimizers, and review some of the latest research and approaches in the field. Attendees will gain a clear understanding of how prompt optimization transforms prompting from brittle craftsmanship into a robust, reproducible, and scientific workflow.

Shashi Jagtap is the Founder of Superagentic AI and a former Apple engineer. He is building the next-generation agentic AI stack from developer tools to orchestration, focused on evaluating, optimizing, and deploying full-stack, production-ready AI agents. He created SuperOptiX, an evaluation-first, optimization-core framework for agentic AI, designed to help companies accelerate their journey into agentic AI systems. Shashi also organizes the London Agentic AI Meetup and the London AI-Native Test Automation & DevOps Meetup, fostering community and knowledge-sharing in the fast-evolving AI ecosystem.

Vibe Prompting: The Secret to Perfect-First-Try Vibe Coding

Ever found yourself doing the "vibe coding" dance? You know, where you have a brilliant idea but end up spending countless hours refactoring because the implementation didn't quite match your vision? What if I told you there's a better way to vibe with AI? Enter "vibe prompting" – the art of having meaningful conversations with AI before diving into code. Instead of immediately jumping into implementation and dealing with multiple refactoring cycles, we'll explore how to craft intentional dialogues that shape better outcomes from the start. Some strategic planning will allow you to use fewer tokens than refactoring your code a hundred times trying to hone in on what you really needed to build in the first place. Ready to take your vibe coding to the next level? Let's explore how "vibe prompting" can transform your development workflow.

I am a long-time open-source developer, educator, manager and mentor. I encourage everyone around me to share their collective knowledge, and am a strong champion of diversity in the tech industry. When I'm not working, I'm helping others with career coaching and interview preparation, tinkering with 3D printing or IoT projects, or telling dad jokes. I'm originally from Canada but currently live in the USA.

Role-Based Prompting for Image Generation: Cultural Precision Through Reasoning

Getting an AI-generated image to capture nuanced, context-specific details—such as authentic cultural references—can be surprisingly difficult. In this session, we’ll explore a practical, reasoning-driven workflow for bridging that gap. The process begins by constructing a detailed persona for a subject-matter expert (for example, a historian of Japanese culture). Using a reasoning engine, we then generate prompts from the expert’s perspective, feed them into an image generation model, and finally, apply AI-based analysis to evaluate the result—closing the loop with metacognitive feedback. I’ll share real-world examples where this approach produced strikingly authentic and context-aware images, along with tips for adapting it to your own creative and professional use cases.

Ajit Jaokar is a Lectu Engineering Sciences at the University of Oxford, where he also serves as Course Director for the "Artificial Intelligence: Cloud and Edge Implementations" program. As an AI Ambassador at Oxford, he leads interdisciplinary AI reskilling initiatives and is actively shaping the future of AI education. Ajit is a Senior AI Fellow at the UK Ministry of Justice’s Justice AI Unit, advising on the national rollout of large-scale AI systems. He is the founder of Erdos Research and feynlabs, and his work focuses on early-stage AI prototypes and democratizing AI through accessible education. Ajit is a frequent speaker at global forums including the World Economic Forum, the European Parliament, and the G7. A passionate advocate for neurodiversity, he is currently writing a book on teaching AI through mathematical foundations to high school students.

The Prompt is Dead, Long Live the Context

In case you didn't know, the era of the prompt as the sole monarch of AI engineering is over. For years, we've treated prompts like magic incantations, but this narrow focus creates brittle, amnesiac systems that fail in the real world. The simple prompt is dead. Its successor? Context Engineering. This session is a practical guide to this new paradigm, showing you how to build truly intelligent, production-ready AI agents using Google's Agent Development Kit. We will move beyond the single prompt and learn to architect the entire information ecosystem—the rich, dynamic context—that allows Gemini to perform complex, multi-step tasks with stunning accuracy. You will leave with a new playbook for building sophisticated AI, including: - A Live Demo in Firebase Studio: We will build and deploy a context-aware sales assistant, demonstrating how to orchestrate Google's powerful Agentic Workspace. - Mastering Retrieval with Vertex AI Search: Learn to ground your agents in reality by dynamically fetching information from your own knowledge bases, eliminating hallucinations. - Interacting with the World via Function Calling: See how to give your agent "hands" by enabling it to interact with external APIs and data sources in real-time. - Deploying for Scale with Vertex AI Agent Engine: Understand the path from a prototype to a scalable, production-grade agent. It's time to stop whispering into a keyhole and start architecting the world behind the door. The prompt had its reign. Now, let us build the future with context and Google Cloud.

Muhammad Ahsan Ayaz is a Google Developers Expert in Angular, an author of 3 books, and a Software Architect at Scania. He has received multiple awards as an educator, created open-sourced projects used by thousands of developers, and created engaging video courses as tutorials which hundreds of developers benefit from, every day.

Unlock Your Team's Potential: Building Your Agency's Prompt Playbook

Are you ready to harness the full power of AI, but not sure where to start? This presentation is for you! We'll demystify prompt engineering and show you how to build a custom prompt library and internal training program right within your own agency or team. Many agencies struggle to consistently leverage AI across all their services. We faced the same challenge and developed a sustainable framework to empower our various teams and roles, including sales, business development, marketing, strategy, campaign management, development and project management. In this session, we'll share our journey, including the pitfalls we encountered, so you can avoid common mistakes. You'll learn practical strategies for: - Crafting effective prompts that get the results you need, with consideration for the job to be done and client needs - Organizing your prompts into a valuable, easy-to-use library, including how to iteratively improve on prompts as a team - Training your team to confidently and consistently use AI tools, including selecting the right model and tool for the job to be done Join us to discover how a well-designed prompting framework can streamline your workflows, enhance client support, and unlock new efficiencies for agencies.

With over 18 years of experience, Richard embodies a key value of RDEL Group: “Explore What’s Possible”. As a Business and Commerce Marketing graduate from the John Molson School of Business at Concordia, Richard is responsible for RDEL's products and operations, including AI. He ensures the goals and objectives of our partners and clients are met by leading RDEL’s AI and paid media teams. Richard's role extends into the oversight of all reporting, technology, and ancillary marketing services that RDEL offers. During Richard’s previous roles with RDEL, he shepherded major changes to the marketing division which resulted in significant growth for the company. At the paid media division, Media Propulsion Laboratory, what was once a three person team has expanded to an 12 person team within the last 2 years with an impressive client roster. The seamless expansion would not have been possible without Richards' keen leadership skills. Richard has secured impressive results for MPL’s clients. For example through-out our nearly 10 year partnership with Xplore, a rural and remote internet service provider, he secured an overall conversion increase of 221%, accompanied by a 43% reduced cost per acquisition (CPA). Prior to joining RDEL, Richard managed and grew accounts at The Canadian Professional Sales Association, AutoTrader, Xplore (formerly Xplornet), and VIA Rail Canada.

Prompt Driven Development (PDD) or the Art of Prompting

AI is like steroids. It can speed up the process, but you still have to go to the gym and practice the fundamentals. You have to suffer. (c) In this talk, I will explore the art and science of writing actionable, high-context prompts that transform AI from a "hallucinating assistant" into a productive teammate, without skipping the basics. We will start with a brief historical overview, comparing today’s Prompt-Driven Development (PDD) to the Readme-Driven Development (RDD) practices of the past. I will present practical techniques for maximizing the quality of AI-generated code by carefully tuning context, specificity, and scope, all demonstrated with concrete and relatable examples. I will also describe the concept of AI wrappers, the essential middleware behind tools like Cursor and Windsurf. Wrappers automatically gather context from the codebase, tailor prompts for optimal results, refine AI output, and embed changes directly inside editors like VS Code. This section will clarify why prompt engineering is a critical modern skill. I will also show my personal prompt collection in action. Then we will break down what makes an effective prompt and what does not, comparing and iterating live. I will illustrate why different scenarios require different types of prompts and explain why this matters for developers at all levels, not just juniors. To wrap up, I will outline key principles for AI-assisted development, including requirements analysis, collaborative habits, validation practices, and robust documentation. Attendees will leave with actionable guidance for integrating prompt-driven practices into their daily workflow. Slides What is PDD The Evolution: From RDD to PDD Prompting Matters: Context, Specificity, Scope My Prompt Collection Why We Need Different Prompts Why Prompting Is Important (not only for juniors) The Golden Rules of Vibe Coding

Platform Engineer | OSS Contributor Vlad is a seasoned full-stack developer with over a decade of experience building and maintaining scalable B2B platforms, as well as a dedicated open-source contributor. He currently works at Cybergizer in a hybrid SWE/CRE/SRE role, focusing on production reliability, systems design, and cross-functional engineering using languages like Ruby, Elixir, and Rust. He is a member of the Diesel.rs contributor team and the creator of opencryptolist.xyz, a platform dedicated to fostering open-source contributions in the blockchain industry. Vlad is also the author of getIDL.xyz – a tool to get IDL and source code info from closed-source Solana programs, using AI-based reverse engineering. He has open-sourced, maintained, and contributed to several libraries in the Ruby and Rust ecosystems (including pkcs12cracker, solscan-mcp, and visual-cryptography). He is also a writer for EffectiveProgrammer, AI Advances, IT Next, and Level Up Coding, and a three-time hackathon winner.

Beyond Prompt Libraries: Teaching AI (and Your Teams) to Ask Better Questions

Prompt libraries are yesterday’s solution. Copy-paste tricks don’t scale and they don’t create resilient teams. In this session, Erik Schwartz will show executives how to move beyond libraries and towards prompt self-sufficiency, using meta-prompting techniques that teach AI (and your people) to ask better questions, adapt on the fly, and build a culture of prompt fluency. Walk away with practical frameworks and meta-prompts you can put to work immediately.

Erik Schwartz is a technology leader, entrepreneur, and founder of theAiExpert.ai, with over 25 years at the intersection of search, knowledge discovery, and artificial intelligence. He has held senior roles at Comcast, Elsevier, and Microsoft, where he pioneered large-scale platforms and introduced generative AI into mission-critical products used by millions. At Elsevier, Erik spearheaded ScopusAI, a breakthrough generative AI solution that transformed how researchers discover and evaluate knowledge — a project that exemplifies his talent for turning cutting-edge innovation into measurable business value. Today, through theAiExpert.ai, Erik acts as a Fractional Chief AI Officer (CAIO), helping organisations move from AI-curious to AI-powered. He designs agentic workflows, compliance accelerators, and adaptive AI systems that cut costs, reclaim time, and unlock growth. A regular speaker, writer, and host of the London AI Dinner, Erik is recognised for his ability to demystify AI for executives, inspire bold thinking, and deliver strategies that bridge experimentation with impact.

Multilingual by Design: Adapting Prompts and Processes Across Learning Languages

This talk lifts the lid on how we have built processes around our new features, while onboarding language experts to become prompt engineers.

You’ll see how we:

• Co-design prompts with subject-matter experts
  Turning teaching strategies into reproducible prompt patterns for pronunciation and conversation feedback.

• Personalise feedback with learner-aware prompts
  Adapting to errors, goals, and CEFR level—while guarding against over-correction and jargon.

• Translate and adapt prompts across languages
  Building a multilingual prompt layer that preserves intent, pedagogy, and tone.

• Make feedback measurable
  Using rubric prompts and scoring agents to assess feature outputs for accuracy, level-fit, clarity, and encouragement—driving a continuous evaluation loop and consistent judgments across languages.

Expect concrete templates, failure modes we hit (and fixed), and the human-in-the-loop practices that kept educational integrity front and center.

If our students are turning to AI to build speaking confidence, this is how teachers and engineers must co-design the prompts—and the evaluators—that truly serve them.

Ilya Kiselev Busuu, Lead Al Product Manager Ilya Kiselev is a Lead Al Product Manager with over seven years of experience building Al-powered products that address real-world challenges. Ilya currently leads Al Product strategy at Busuu (a Chegg company). There Ilya spearheaded initiatives that power and enhance the learner experience, ranging from LLM-based speaking practice to learner assessment using more conventional ML models, fostering user growth and engagement. His career spans work in computer vision, semantic search, and user research at companies like Satis.ai, Attraqt, and Elastic. A former startup founder and published neuroscience researcher, Ilya blends technical expertise with a passion for user-centered Al innovation.

Keynote: Beyond Output: Evaluating Agentic Workflows in LLM Systems

As LLMs evolve into multi-step agents, evaluation must move beyond input-output correctness. This session shows how to assess both node-level decisions and full agentic flows, with a live demo of multi-node agents and practical eval techniques.

Designing Reliable Agent Workflows with Context Engineering

Building AI agents is no longer the hard part—making them reliable is. This talk dives into the under-discussed but critical practice of context engineering: the techniques used to control what your agents “see,” “remember,” and respond to. The session will examine how to design effective context flows across planning, execution, and memory stages using frameworks such as LangGraph and Google’s Agent Development Kit (ADK). Through real-world examples and a demo, we’ll explore how to handle context drift, token overflow, and prompt injection. I’ll also share a before-and-after demo of a flaky agent that becomes robust simply by improving its context strategy. This session will equip attendees (whether they’re building single-agent copilots or complex multi-agent systems) with practical tools to debug, refine, and productionize agent behavior through clean, modular context pipelines.

I'm a Tech Lead at Red Buffer, where I design and deploy generative AI systems for real-world use—ranging from healthcare automation to multi-agent copilots for SaaS platforms. I'm also a Google Developer Expert (GDE) in AI, with a focus on making advanced LLM systems accessible, robust, and production-ready. When I'm not leading AI projects or running workshops, I mentor learners at Turing College as a Senior Team Lead, helping them bridge the gap between theory and real-world engineering. I also write for Towards Data Science, where I break down complex AI topics with a focus on practical use cases. Outside of tech, I have a deep love for classic literature (yes, Tolstoy counts—even if he's Russian), strong coffee, and the quiet company of cats. I'm driven by curiosity, clarity, and the joy of building things that make life a little smarter and simpler.

Getting started with Chrome Built-in AI APIs

AI offers a wide array of applications, from audio transcription and image description to writing assistance, proofreading, content translation, and diverse general prompting purposes. For implementing these kinds of use cases, you broadly have three options: - Cloud solutions – send your data to an AI model running on a powerful server in the cloud - On-device solutions – download a (possibly large) model file to the client and run it locally - Shared models – use models built into a device or browser (the focus of this talk) On the Chrome team, we’re proposing a number of APIs that we and partners we’ve worked with think would be worthwhile to have as common Web platform APIs. In this talk, I’ll: - Give an overview of these APIs - Briefly demo some implementations in Chrome - Open the floor for general discussion and feedback

Thierno currently works as a Web Ecosystem Engineer @Google focusing on on integrating Gemini into Google Chrome and making Web AI accessible for everyone.

Meta Agents: Self-Building Al Agents in a Vibe-Driven World

Discover how Meta Agents transform natural-language instructions into fully operational Al agents. In this session, we'll explore the "vibe-driven" approach to architecting Al Agents - where you describe the Al Agent you need, and a Meta Agent designs, builds, and deploys it for you. The Meta Agent captures your requirements, organizes the right tools and knowledge bases, defines the system prompt, and handles AWS deployment. Participants will see a live demonstration in which we will interact with a Meta Agent to create a new Al agent, and we will test it together. We will unpack the design decisions, methodology, and process, leaving the audience with practical skills to architect and deploy their own version of Meta Agents.

Carlo Peluso is a Senior Solutions Architect at Storm Reply, focusing on Generative AI solutions and cloud-native data architectures. Since 2022, he has been supporting organizations in designing and scaling AI-driven and data platforms on the cloud. He is also an AWS Community Builder and a frequent speaker at conferences and tech events on Generative AI. Guido Nebiolo is a seasoned IT professional with over 10 years of experience and a strong background in cloud architectures, software engineering, and artificial intelligence. Having held various leadership and technical roles, he is currently a BU Manager at Storm AI Reply, a leading cloud and AI consulting company. In this role, he oversees strategic AI initiatives and leads a specialized team delivering Generative AI solutions. Guido helps organizations scale their AI capabilities through both technical implementation and organizational change management, serving clients across multiple industries. He is also an AWS Community Builder and AWS User Group Leader for AWS Turin (Italy) UG, and an active speaker at multiple conferences and events.

Context Engineering in LLM Pipelines

In this talk we'll walk the audience through the modern AI-powered service set up, from the very basics to technicalities of hosting AI applications in a cloud. We will describe how to convert the idea into the AI-powered pipeline and then what challenges awaits you when rendering this theoretical concept into live. We will talk about: * Structured inputs and outputs and how engineer the context * How to host such an application in a cloud and what building blocks will you need * The importance of observing and monitoring the performance: what problems to expect and how to battle them * The criticality of avoiding vendor lock-in and the need to experiment quickly with models and prompts

Sergey Konstantinov is an experienced web developer with over 15 years in the field, from heading a 50M DAU API service to working as a tech lead in a startup. He is particularly interested in APIs as a broad topic, about which he has given many speeches and authored two books. Alena Panshina is a product manager and consultant with 13 years in tech. Former Yandex PM and ex-CEO of a $2M AI startup, she has been working with LLMs since 2019. She has scaled products from zero to millions of users and now helps companies build AI-powered solutions. Based in Ireland, she has traveled to 40 countries.

Prompting LLMs to use Tools and Resources with MCP

The majority of Model Context Protocol (MCP) use right now is directly through end user applications like Claude Desktop or Cursor. However, pretty much any Large language Model, from small open models, to the APIs of the frontier models can be prompted to call tools and retrieve external resources, including those exposed using the Model Context Protocol (MCP). Doing this effectively in your projects is key to the development of agentic software. This talk will explore how different models respond to different types of tool use prompting, along with the benefits and pitfalls prompting to generate tool calls in different data formats. We will go through some real examples of interacting with MCP servers, and look at the prompts used by some popular agentic frameworks and tools.

Experienced software professional focussed on Machine Learning and Apps with a passion for Open Source AI and a strong interest in novel ways Machine Learning can be used on locally on low power devices. In 2024 Richard delivered a talk about using the open source CodeGemma model for Google DevFest Scotland, and spoke about the current state of Open Source AI at the Macc Tech event. Richard is the CTO at Blue Beck Ltd, and has a background in backend service and mobile development, going back to when the first releases of the Android and iOS SDKs became available, on other mobile devices before that, having started his career at RARE as a console game developer.

Prompt Injections, Red Teams, and the Unsolvable Problem

Join Sander Schulhoff, creator of the first prompt engineering guide and leader in AI red teaming, for a deep dive into the evolving world of AI security. This session explores the challenges of defending against prompt injection attacks, the realities of adversarial robustness, and the future of prompt engineering. With stories from major AI labs, practical prompting techniques, and candid discussion on the unsolved problems in AI safety, this conversation is a must-watch for anyone interested in the cutting edge of artificial intelligence.

Sander Schulhoff is the CEO of InventoryQuant and Learn Prompting. He created the first Prompt Engineering guide on the internet, two months before ChatGPT was released, which has taught 3 million people how to prompt ChatGPT. He also partnered with OpenAI to run the first AI Red-Teaming competition, HackAPrompt, which was 2x larger than the White House's subsequent AI Red-Teaming competition. Sander's background is in Natural Language Processing and deep reinforcement learning. He recently led the team behind The Prompt Report, the most comprehensive study of prompt engineering ever done. This 76-page survey, co-authored with OpenAI, Microsoft, Google, Princeton, Stanford, and other leading institutions, analyzed 1,500+ academic papers and covered 200+ prompting techniques.

Memory Engineering: Going Beyond Context Engineering

Large Language Models (LLMs) are powerful, but their limitations become clear in multi-turn interactions: they lose track of context, repeat mistakes, and forget what matters. Lately, developers have relied on context engineering—clever prompt design, retrieval pipelines, and compression—to work around these constraints. But context alone is ephemeral. To build agents that are reliable, believable, and capable, we need to move beyond context and into memory engineering.

This talk introduces memory engineering as the natural progression of context engineering, exploring how to design systems where data is intentionally transformed into persistent, structured memory that agents can learn from, recall, and adapt with over time. We’ll walk through the data→memory pipeline, types of agent memory (short-term, long-term, shared), and practical strategies like reflection, consolidation, and managed forgetting.

Finally, we extend the conversation with a Context Engineering++ perspective—a holistic view of how memory, context, and attention can be engineered together to enable the next generation of agentic systems. Attendees will leave with a clear framework for evolving from prompt engineering to context engineering to memory engineering, and practical guidance on how to architect agents that don’t just respond, but remember, adapt, and grow

Richmond Alake is an AI strategist and engineer specializing in the emerging discipline of memory engineering for AI agents. At MongoDB, he leads initiatives at the intersection of data, developer experience, and generative AI, driving adoption of AI-native architectures and agentic systems.

As the creator of Memorizz, an open-source framework for building memory-augmented AI agents, Richmond has been at the forefront of defining how persistent memory transforms LLMs from stateless chatbots into adaptive, evolving systems. His thought leadership—spanning conference talks, technical cookbooks, and industry publications—focuses on bridging research insights with practical engineering patterns for building reliable, believable, and capable AI.

He has written for leading publications including NVIDIA, Neptune AI, and O’Reilly, authored over 200 articles on AI and developer experience, and collaborated with Andrew Ng on a Retrieval-Augmented Generation (RAG) course for DeepLearning.AI.

Through his work, Richmond helps organizations and developers alike navigate the shift from prompt engineering to context engineering to memory engineering, charting a path toward the next generation of agentic AI.

Logic Meets Automation: How DSPy and Prolog-Style Reasoning Boosted GPT-3.5 Accuracy by 19.5%

Large Language Models (LLMs) often fail not because they lack capability, but because the prompts guiding them are inefficient. Manual prompt engineering is slow, inconsistent, and nearly impossible to scale. In this talk, I’ll share how I tackled this in my open-source project, prompt-eng-gsm8k-gpt3.5-dspy, where I used DSPy to automate prompt creation and tuning for the GSM8K mathematical reasoning benchmark. By defining tasks declaratively in DSPy and letting its compiler handle generation and refinement, I compared zero-shot (55.0%), few-shot (60.5%), chain-of-thought (68.0%), self-consistency (72.5%), Prolog-style (71.0%), and my Enhanced Prolog method — which achieved 74.5% accuracy, a 19.5% gain over zero-shot and 6.5% over CoT, while using less time and tokens than self-consistency. This Enhanced Prolog approach structures the LLM’s reasoning as logical facts and inference rules, making outputs: • More accurate by avoiding reasoning leaps • Debuggable via traceable steps • Machine-verifiable through logical consistency checks Attendees will learn: 1. How DSPy automates and scales prompt optimization. 2. When to use zero-shot, CoT, few-shot, self-consistency, or logic-based prompting. 3. How Prolog-style reasoning improves reliability and explainability. 4. Setting up A/B testing pipelines for prompt evaluation. 5. Debugging workflows for more trustworthy LLM outputs. Packed with real metrics, open-source code, and a reproducible workflow, this talk moves you from guesswork to data-driven prompt engineering that’s explainable, efficient, and production-ready.

I am a Master’s student in Artificial Intelligence at Royal Holloway, University of London, specializing in prompt engineering, LLM optimization, and NLP. I created the open-source project prompt-eng-gsm8k-gpt3.5-dspy, where I used DSPy to automate prompt creation for the GSM8K benchmark and developed my Enhanced Prolog-style reasoning method, boosting GPT-3.5 accuracy from 55.0% (zero-shot) to 74.5%. I have hands-on experience testing prompt strategies — zero-shot, few-shot, chain-of-thought, self-consistency, Prolog-style, and DSPy-optimized — across multiple models, including GPT-3.5, LLaMA-2-7B-Chat-HF, and Qwen. My work combines reproducible code, measurable results, and practical workflows that bridge research and real-world applications. Outside AI, I enjoy doodling, playing badminton, and mentoring peers in AI projects. And while I don’t have a definitive favorite member of One Direction, I’ll say Harry Styles for his creativity — a trait I also bring to my AI experiments.

From GraphQL to MCP

The Model Context Protocol (MCP) is a great base protocol that allows AI agents to connect to 3rd party systems, and has gained rapid adoption across the industry. However, it requires designing APIs with specific patterns that make them easy for AI agents to use. Your organisation likely already has pre-existing APIs with machine readable schemas, like GraphQL or OpenAPI (REST). While they expose useful business data and capabilities, these schemas are not optimised for consumption by AI agents. Still, they have mature frameworks around them and importantly – are already implemented. In this talk we'll share what we learned at Isometric when building a translation layer between GraphQL and MCP. We optimised for avoiding duplicate work and making the most of existing tooling built into the GraphQL server, while at the same time ensuring that the end result follows all the MCP server design best practices and can be effectively used by AI agents. Our internal GraphQL server has more than 800 types, 100 top-level query fields and 200 mutations. We'll show how we leveraged GraphQL schema introspection, query templates and complexity heuristics to distill this large API surface into a small, but scalable, set of MCP tools. You’ll learn about practical strategies for exposing your existing APIs to AI agents without rewriting the entire API layer. We'll discuss specific challenges like handling GraphQL's infinite nesting problem and demonstrate how complex GraphQL queries get transformed into simpler, AI-friendly MCP tools.

Konrad Komorowski is a software engineer at Isometric, working on carbon removal credit verification. He's spent his career between big tech and startups – building probabilistic ads measurement systems as a tech lead at Meta, and helping scale engineering organisations at companies like Glovo from 150 to 500 engineers. He's passionate about finding simple & robust solutions to complex, real-world problems, believing the best engineering happens where theory meets practice.

Peteris Bikis is a Staff Software Engineer and Product Leader at Isometric, building tools to advance carbon removal credit verification. Previously, he was Principal Frontend Engineer at Packfleet, helping scale its logistics platform, and Principal Engineer at Pollen, where he led engineering across consumer and internal products. Earlier in his career, he co-founded the award-winning interactive studio Asketic, and worked at Citymapper shaping the future of urban mobility. With a background spanning design and engineering, Peteris is passionate about creating elegant, user-focused systems that bridge technical depth with product vision.

Lessons learned from teaching & building AI Second Brain

Over 18 months, JarvisJr evolved from a naive “search everything” chatbot into an intelligent memory agent that decides when to search.

Built around the CODE framework (Capture, Organize, Distill, Express), it transformed from technically impressive but contextually inefficient into genuinely useful.

Results: - 3× faster responses
- 80% cost reduction
- Actual context understanding

Goda Go and Sjoerd Tiemensma share hard-won lessons from building an AI Second Brain and onboarding 200+ community members along the way:

• Why “smart context” beats traditional RAG—AI that thinks before it searches
• The breakthrough that cut setup time from 60 minutes to under 10
• Scaling & modular system design that adapts from one user to supporting teams with privacy and access controls
• How MCP and Supabase Edge Functions transformed the architecture
• The hardest problem: teaching AI when to forget, not just remember
• What users actually requested: control over magic, not more magic ```

Sjoerd Tiemensma is an AI engineer and automation specialist based in Arnhem, Netherlands. Without any coding background initially, Sjoerd built a career creating practical AI solutions for businesses, specializing in automations and agent systems that deliver real value. He builds custom AI tools, including a multi-agent marketing platform featured on Dutch television. Sjoerd runs the "Use AI" newsletter with over 2,000 readers, writes for AI Report (the largest Dutch AI publication), and co-runs the AI Productivity Hub community with Goda, where they teach practical automation and help people implement AI tools that actually save time.

Goda Go is a YouTube creator (100K+ subscribers), TEDx speaker, and co-organizer of the Prompt Engineering Conference. Based in Berlin, she co-founded the AI Productivity Hub with Sjoerd Tiemensma to help non-technical business owners and professionals implement AI without coding. Goda trives to transform how solopreneurs use AI for productivity by creating scalable & secure modular memory systems. Her background spans architecture, product marketing, and entrepreneurship, bringing a unique perspective to making AI practical and accessible. She specializes in AI automation, memory systems, and teaching business owners to move from overwhelmed by AI tools to confidently implementing them.

Keynote: Beyond Output: Evaluating Agentic Workflows in LLM Systems

As LLMs evolve into multi-step agents, evaluation must move beyond input-output correctness. This session shows how to assess both node-level decisions and full agentic flows, with a live demo of multi-node agents and practical eval techniques.

Can AI Develop a Whole Videogame?

From prompt to play: witness how cutting-edge generative models, clever chaining and a sprinkle of game-dev know-how can birth an end-to-end, fully-playable beat ’em up... Including art, music, code and lore.... just in my spare time. Get inspired, steal workflows, ship your dream game.

Jesus has 30 years of experience in the digital sector, including 19 years at Microsoft, where he specialized in blending Artificial Intelligence and User Experience, drawing on his background as both an engineer and designer. In 2014, he became the inventor of a protected Microsoft patent, earning recognition for his innovative work.Jesus has collaborated with over 200 clients across 35 countries in industries like sports, entertainment, media, and retail. He has led the design and development of consumer apps with millions of downloads and is an award-winning speaker with over 100 sessions and keynotes at global events. He also serves as an AI Tutor and Course Co-Director at the University of Oxford.

Algorithmic Context Engineering for Agents with Optimizers

Could we use algorithms to help solve our next bottleneck? As LLMs become multi-step agents baked with tools and varying degrees of complexity improving how context is handled from cost, latency to accuracy becomes a compounding problem: shaping what agents see, remember, and do across tools, memory, and retrieval while continuously improving the underlying ai application without manual prompt surgery.

Jacques is Comet’s Head of Product. A data scientist and ML practitioner at heart, served in data science roles at Triptease, Worldpay, and Brightmile. He founded Stakion, a platform Comet later acquired, which monitored machine learning models in production for breaking changes and concept drift. He holds his master’s degree from Imperial College in London, and his engineering degree from CentraleSupélec in île-de-france.

Hacking the Authoring Process: Lessons from Writing a Book with AI

Over the last 18 months, I didn't just write a book about the AI revolution; I wrote a book with it. My book, From Cloud Native to AI Native, Catching the Next Wave of Innovation, was created in a deep, evolving partnership with generative AI. This talk is the story of our collaboration. I will take you from my initial, tentative prompts to the complex dialogues that shaped entire chapters, sharing real examples of prompts and their raw outputs. We'll explore how my own skills had to evolve alongside the AI's capabilities. This journey provides a powerful parallel to the core message of my book: becoming "AI native" isn't about replacing human skill but augmenting it. You will leave with a clear-eyed view of how to move beyond simple queries and integrate LLMs into any serious, professional workflow.

Prompt Your Way to Calm: Using AI to Reduce Stress and Avoid Burnout

Even the most skilled developers can hit a wall—stress and burnout silently sabotage focus, creativity, and productivity. What if the same AI tools you use to write code could also coach your mind to stay calm, clear, and resilient? In this talk, I’ll show how to design prompts that go beyond outputs—transforming Large Language Models into adaptive AI “coaches” for wellbeing. Drawing on clinical hypnotherapy, life coaching, business psychology, and my experience as a Java developer who experienced burnout firsthand, I’ll share how prompts can: Reduce cognitive overload and decision fatigue Reframe stressful challenges into actionable steps Create AI-guided daily reset rituals Maintain sustainable focus without sacrificing creativity. You’ll walk away with ready-to-use prompt templates that transform AI into a proactive wellbeing partner—helping you not just survive, but thrive in high-pressure tech roles.

Gena Frangina is a Clinical Hypnotherapist, Software Engineer (Java), IT Wellbeing Strategist, and Mindvalley HoloBody Coach. After experiencing a panic attack while coding during the pandemic, she dedicated herself to mastering hypnotherapy and coaching strategies for preventing burnout. Today, she blends hypnotherapy, coaching, and software engineering expertise to help tech professionals use AI not just for productivity, but as personal wellbeing partners—empowering them to build resilience, focus, and sustainable performance in high-pressure environments.

Ai Agents - Fireside Chat

Translating Prompts, Transforming Communities

A simple Pull Request to translate the Prompt Engineering Guide sparked a journey—from a late-night side project to mentoring in OSSCA’s LLM program. This talk shares how curiosity, community, and open source shaped a growing AI movement in Korea.

Coding, Rewritten: What 10.000+ AI-Generated Code Completions Taught Me About the Future of Software Development

What if coding wasn't just about writing code anymore? After a year-long deep dive into over 30 AI coding tools — including ChatGPT, Claude, Google Gemini, DeepSeek, Grok, GitHub Copilot, Cursor, Windsurf, Claude Code, Lovable, V0, and many… more — I’ve learned how radically the developer experience is evolving. As a former full-stack (Java) engineer turned Prompt Engineering Advocate, I stepped away from the keyboard and let AI agents take the lead. But I also coded together with AI. With more than 5,000 AI-generated code completions only on Windsurf, I’ve seen what works, what doesn’t, and what’s coming next.In this talk, I’ll reveal what happens when AI agents don’t just complete your code — they challenge your logic, suggest UX improvements, plan, and autonomously refactor your functions. You’ll learn how responsibilities shift in agent-amplified workflows, how to manage memory and context windows, what Flush Coding is, what rule and planning files are, why our refactoring mindset makes it hard to adapt, agentic swarms and why not VibeCoding but AmpCoding might just be your next paradigm. We’ll walk through real examples, transforming a screenshot into a working application and debugging with agents that understand your enterprise codebase. You’ll leave with practical insights, clear takeaways, and a vision of how to stay in control — even when you’re not the one coding. You will receive a report out with community-driven insights. It’s still your name on the commit. Let’s make it count.

Jeroen Egelmeers is a Prompt Engineering Advocate and GenAI Whisperer at Sogeti Netherlands. He also serves as a Software Engineering Trainer at the Capgemini Academy and is one of the authors of the latest TMAP book. Jeroen frequently delivers presentations on Prompt Engineering, AmpCoding, and Software Quality at various events and venues. He is the founder of the Crafting AI Prompts Framework — a framework designed to help create optimal prompts while addressing concerns such as non-disclosure, data security, GDPR compliance, and other constraints. He also tracks the latest tools, models, and strategies for Agentic Programming on the framework's website.

Prompting Your Way to a Search and Recommender System

Can prompting and context engineering be enough to build a production-ready search and recommender system? We compare keyword, vector and hybrid retrieval methods, then introduce Superlinked’s mixture-of-encoders architecture, which combines specialized encoders with LLM query understanding for better relevance. With AI-driven coding workflows, we show how prompting can scaffold components, guide evaluation and accelerate iteration. Because Superlinked provides the structure and resources to handle multi-attribute and metadata-aware queries out of the box, teams can move from prototype to production much faster. We also cover integrations with LangChain and LlamaIndex, making it easy to connect with existing pipelines. Attendees will learn practical strategies to build high-quality search and recommender systems quickly using prompting and the open source Superlinked library.

Filip Makraduli is a machine learning engineer with a strong background in AI systems, vector search, and large language models (LLMs). He holds a Master’s degree in Biomedical Data Science from Imperial College London. Currently, Filip works as a founding developer relations engineer at Superlinked, where he focuses on building real-time, multi-attribute search and recommendation systems. His work emphasizes the use of multi-encoder architectures to enhance retrieval quality and reduce reliance on reranking strategies. In the past, Filip worked as a data scientist at Marks & Spencer, where he contributed to AI-driven solutions for retail. He has also held machine learning engineering roles across several UK-based startups, focusing on applied AI and product-oriented ML development. In addition to his industry work, Filip has been active in the open-source community, particularly around LLM tooling and pipelines. He has delivered various talks on practical machine learning applications.

AI, Prompting and Mental Health

This session explores the emerging role of artificial intelligence (AI) in mental health, considering its application both within mental health services and everyday use by the public. AI is already being trialled in areas such as digital triage, clinical decision support, mood monitoring, and therapy-adjacent tools, raising important questions about efficacy, safety, and integration with professional practice and the UK regulator NICE is appraising two interventions with LLM or AI components.

Alongside these developments, the ways in which lay people engage with AI also merit attention. Through everyday prompting, users can elicit supportive reflections or structure wellbeing routines, but poorly framed question alongside generic model training and the absence of "safety rails" may produce misleading or unsafe outputs or delay seeking professional help when necessary.

The session will situate these practices within broader debates about ethics and governance. Key issues include clinical reliability, data privacy, cultural bias, and the boundary between self-help tools and regulated interventions. By examining both institutional and informal uses of AI, this session highlights the promise of scalable, accessible support, while emphasising the need for robust evidence, regulation, and ethical safeguards in this rapidly evolving field.

Dr David Crepaz-Keay, FRSPH, is Head of Research and Applied Learning at the Mental Health Foundation, a 75 year old public mental health NGO. He leads the London based research team and is responsible AI, LLMs and knowledge management systems for the organisation.

Dr Crepaz-Keay is a former co-chair and now member of the ethics, policy and position committee of the International Society for Psychiatric Genetics and a fellow of the Royal Society for Public Health. He is an editor of the Journal of Mental Health Training, Education and Practice; the Oxford Handbook of Philosophy and Public Mental Health and the Handbook of Phenomenology, Values and Clinical Decision-Making in Personalised Mental Health Care. He has been a technical advisor for the World Health Organisation, senior mental health advisor to Public Health England, written a mental health module for the Open University, written, spoken and campaigned for improvements in mental health services.

Orchestrating the Next Billion Agents

Prompt engineering has taken us far, but it’s reaching its limits. Current AI systems are fragile: prompts are brittle, agents are isolated, and orchestration often collapses into spaghetti glue code. To unlock the next leap, we need to move beyond tricks and hacks and start thinking in systems.

This talk explores a graph-native approach to reasoning and orchestration: - Knowledge Hypergraphs as context engines that capture relations, causality, and logic — the missing symbolic backbone for LLMs. - A layered reasoning model that navigates from atomic concepts through compositional reasoning, contextual integration, meta-reasoning, and finally external knowledge integration. - Graph-native multi-agent orchestration, where agents collaborate using strategies that mirror real-world organizations: sequential pipelines, hierarchical teams, debates, juries, and emergent coordination. - Agentic governance policies that provide enterprises with what they actually need: compliance, auditability, and trust.

Instead of chaining prompts, we get composable reasoning: agents dynamically choosing strategies, grounding their outputs in knowledge graphs, and collaborating under policy-driven rules.

The talk closes with a glimpse into how these ideas are being implemented in practice. I am actively building OneSys, an open-core AI operating system that embodies this architecture: the hidden conductor for the next billion AI agents.

Mattis Stene-Johansen is the founder and CEO of Rhizomatic Labs AS, a software lab building OneSys, where he is pioneering the development of an AI operating system designed to orchestrate the next billion agents.

At just 22, Mattis is a young and passionate technologist who has spent nearly half his life programming. He combines hard work, curiosity, and systems-level thinking with a deep love for his craft.

He holds a B.Sc. in Computer Science from the University of Oslo, is a Google Certified Cyber Security Professional, and is an Ex-Ethereum contributor. His background spans startups, enterprise engineering, and open-source, with a focus on building resilient, privacy-first technology.