In the new AI-enabled coding environment, vibe-coding is both an enticing and often-misunderstood reality.  Let's spend some time with startup experts who have evaluated hundreds of MVPs and coders who can help sort fact from fiction, and position vibe-coding to be a resource for both technical and non-technical founders.

The explosion in agentic AI frameworks like LangGraph, AutoGen, and CrewAI has unlocked incredible possibilities for automating complex tasks. However, this revolution has been overwhelmingly centered in the Python ecosystem, leaving a massive community of skilled mobile, backend, and JVM developers on the side-lines or forcing them into a language they don't use daily. This session introduces Koog by JetBrains, a powerful framework that brings agentic AI development to Kotlin. We will explore how developers can build sophisticated, multi-agent solutions without writing a single line of Python. We'll directly compare Koog's architecture and capabilities for creating agentic workflows against its popular Python counterparts, highlighting the similarities in power and the differences in approach. The true power of this stack lies in its deployability. We will demonstrate how, by leveraging Kotlin Multiplatform (KMP), you can write your AI agent logic once and seamlessly integrate it anywhere: as a backend service with Ktor, directly within an Android app, or even in the browser with Kotlin/JS. Attendees will leave this talk with a clear understanding of how to build their own AI agents using the tools they already know, ready to integrate intelligent solutions across their entire stack.

Agentic AI is racing into prod—often without a safety envelope. This session shows how to ship micro-LLMs and MCP-based agents that are actually verifiable: zero-trust boundaries, policy-as-code guardrails, signed tool invocations, prompt-injection defenses, evaluation harnesses, and audit-ready provenance. We’ll dissect the top failure modes (unsafe tool use, data exfil via prompts, RAG leakage, jailbreaks) and map them to concrete countermeasures: least-privilege tool adapters, contract tests for tools, adversarial evals in CI, and immutable event logs for post-mortems. Expect a gritty, pattern-driven discussion: reference architectures, checklists, and “red team to fix” playbooks that teams can lift on Monday. If you care about regulated stacks or multi-tenant safety, this panel is your north star for provable agent safety.

Abstract: Cloud costs are spiraling out of control for most organizations, with poor visibility, inconsistent tagging, and manual optimization processes leaving money on the table every month. This talk introduces a new paradigm: autonomous AI agents that continuously monitor, analyze, and optimize your cloud spending

What You'll Learn: Discover how to build and deploy AI agents that transform cloud cost management from a reactive, manual process into a proactive, intelligent system. We'll explore real-world implementations of agents that automatically unify data across multiple cloud providers, standardize resource tagging, create cost-to-business mappings, and execute optimization recommendations—all while learning from your organization's unique patterns and constraints.

Key Takeaways: Architecture patterns for building cloud cost management AI agents. Data unification strategies across AWS, Azure, GCP, and hybrid environments Automated tagging and resource categorization using ML techniques. Real-time rightsizing algorithms that balance performance and cost. Implementation of guardrails and approval workflows for autonomous actions. ROI metrics and success stories from production deployments

Target Audience: DevOps engineers, cloud architects, platform engineers, and engineering leaders looking to reduce cloud costs while maintaining system reliability and developer productivity.

We turned academic search into a conversation by designing, building, and deploying a production GenAI Research Assistant that lets students, professors, and researchers ask natural questions about the papers they find and receive grounded answers, summaries, and related content. This talk walks through the full lifecycle from domain grounding and corpus preparation to retrieval augmented generation, LLM selection and adaptation, and prompt plus context strategies that reduce hallucinations while preserving a great user experience. We will cover the architecture behind scalable pipelines, hybrid use of embeddings and multiple models, why and when we combine providers, and the practical trade-offs among cost, latency, and accuracy. Because trust is essential in scholarly workflows, we will dive into privacy and governance for sensitive content, human in the loop review, and the telemetry that drives iterative quality gains and adoption. Finally, we will share our automated evaluation framework that blends overlap and semantic metrics with groundedness checks and rubric based judging, including Databricks MLFlow, and LLM-as-a-judge, to ensure faithful, high quality results. Attendees will leave with concrete patterns for designing, evaluating, and operating GenAI applications at scale in real production environments.

Tired of AI agent theory? Ready to actually build something? In this interactive session, we'll forget the slides and jump straight into the code. For one hour, you're not an attendee, you're an AI architect. I'll guide you through a rapid prototyping sprint using the powerful MCP framework and ADK Toolkit. You will learn by doing, and you will succeed. What You'll Walk Away With: A repeatable blueprint for prototyping AI agents. A working agent, built by you, ready to be customized. The confidence to turn your own AI ideas into reality. This isn't a lecture. It's a launchpad. Let's build.

Prerequisites: None! This session is beginner-friendly. All you need to bring is your curiosity.

Tired of AI agent theory? Ready to actually build something? In this interactive session, we'll forget the slides and jump straight into the code. For one hour, you're not an attendee, you're an AI architect. I'll guide you through a rapid prototyping sprint using the powerful MCP framework and ADK Toolkit. You will learn by doing, and you will succeed. What You'll Walk Away With: A repeatable blueprint for prototyping AI agents. A working agent, built by you, ready to be customized. The confidence to turn your own AI ideas into reality. This isn't a lecture. It's a launchpad. Let's build.

Prerequisites: None! This session is beginner-friendly. All you need to bring is your curiosity.

AI tools are powerful, but they’re not always fair. They can amplify bias, produce fragile outputs, and mislead us with confident but incorrect answers. In this session, we’ll explore how developers can outsmart the AI by designing prompts that reveal these blind spots and by building safeguards into their workflows.Through real-world demos and interactive examples, we’ll see how small prompt changes create drastically different results and examine what this teaches us about bias, responsibility, and trust. You’ll walk away with practical strategies to design AI applications that are not just clever, but fair, transparent, and resilient.Whether you’re experimenting with Google’s AI tools for the first time or already building advanced workflows, this session will give you frameworks and inspiration to approach AI development more thoughtfully.

Migrating a large front-end application, say, moving from AngularJS to React can feel overwhelming at first. It’s easy to think of it as a huge “all-or-nothing” rewrite. But in reality, a gradual migration, done one piece at a time, is not only possible, it's actually the smarter way to go.In this workshop, we'll walk you through what it really takes to pull off a migration like this without disrupting your team or your product. We’ll look at some of the common challenges you’ll run into like dealing with two different routing systems, keeping state consistent across frameworks, and managing bundle size when both frameworks need to live side by side for a while.We'll wrap up by getting hands-on with practical strategies that have worked well in real-world projects, including patterns like the Strangler Fig. The goal is to provide a realistic and replicable path forward so you can modernize your app, reduce tech debt, and end up with a cleaner, more maintainable codebase all without grinding your velocity.

This talk demystifies Kotlin Multiplatform (KMP) project configuration for sharing business logic. Building mobile apps can be complex, and KMP can add another layer of complexity to an already challenging Gradle build setup. We'll dive deep into distributing shared code to iOS developers, and provide tips for making your KMP builds understandable, shareable, and fast.

In the movie Taken, what if Liam Neeson was contacted by AI instead of his Daughter? With new innovations in AI, it’s now possible to replicate the voice of loved ones and influential figures with minimal sample data sets. Combined with the fact that most people have been in at least 1 recorded zoom meeting, this is a ticking time bomb that is soon to ignite. This session will explore real world scenarios where AI impersonated individuals so clearly that relatives panicked and fell victim to extortion scams. We will look at some examples of this happening in the past, tools used to do this, and discuss how blockchain technology can be used to help defend against these types of attacks.

As third-party cookie blocking within the browser becomes widespread, cross-site tracking techniques have adapted in kind. Many malicious actors now rely on navigational tracking, turning a simple link click into a momentary redirect covertly exfiltrates user data. This talk provides a technical breakdown of modern navigational tracking techniques, such as bounce tracking and link decoration. We will then cover how mitigations were designed, standardized, and ultimately shipped in Chromium to address these issues, offering a look into how modern browsers are working to protect user privacy on the web.

AI can be a brilliant aid but also hilariously wrong. In this interactive session, we’ll put AI through a set of real-world (and slightly quirky) prompts to see where it shines and where bias sneaks in. From unexpected stereotypes to fragile logic, you’ll see why even “smart” systems still need careful guidance. Together, we’ll unpack what went wrong, design better prompts, and talk about how developers can keep AI playful and responsible. You’ll leave with a few practical tricks for spotting bias, improving outputs, and building AI that’s not just powerful but fair and fun to use.

Model Context Protocol (MCP) is an open protocol specifying how to provide context to AI models released by Anthropic in 2024 that has become a leading method for integrating tools and information with AI services like ChatGPT, and Claude.  This talk will cover the basics of MCP, explore the benefits, risks, and costs of the protocol and ecosystem, and demonstrate building a custom connector to integrate meaningful data with an AI service before finishing with the two biggest challenges facing MCP (and AI integration in general): permissions and security.

Presentation will be around Reliability in the era of AI.

This session will provide technical teams with practical guidance on embedding compliance into system architecture and development workflows from the outset. Rather than treating compliance as a late-stage hurdle, we’ll explore how early design decisions (such as access controls, audit logging, integration management, and documentation) can significantly streamline future certification processes, reduce costly rework, and support scalable growth. Attendees will gain actionable insights to build systems that are audit-ready by design, enabling faster time to market, stronger customer trust, and long-term operational efficiency.

Vibe Code 101 is an immersive workshop designed to introduce participants to the innovative world of Vibe Coding — a method that combines AI-powered tools with intuitive no-code development. Attendees will learn how to leverage artificial intelligence to turn raw ideas into structured product roadmaps, prototypes, and applications with minimal coding experience. This session will showcase how to guide AI assistants through a conversational process to design, refine, and build applications quickly with a PM approach.

Unleashing Advanced Video Experiences: Integrating the next-gen video codecs and modern HDR formats in Android Multimedia: Discover next generation video codecs for ultra-high quality capture and how to seamlessly implement HDR, HDR10, and Dolby Vision in Android multimedia apps for immersive in-app experiences. This session highlights modern codec’s advances in compression and color fidelity, and offers practical guidance on integrating advanced HDR formats, optimizing performance with Qualcomm hardware, and ensuring robust playback across devices. Perfect for developers aiming to deliver cinematic-quality video and future-ready app experiences using Qualcomm’s leading-edge platforms.

Development is at a huge shift. GenAI, agentic systems, and new coding tools are not just changing how we write code, they are redefining the entire paradigm for how we approach DevOps and manage software development. AI is removing some of the bottlenecks that shaped our current processes, making it possible to move from an idea to a working prototype in hours instead of weeks. This speed changes how we plan projects, how we allocate resources, and how we think about the roles on our teams. This is about building AI-enabled teams and workflows, not just adding AI into our toolchain. In an AI native world, developers, architects, and designers must learn to orchestrate workflows with AI, manage and curate the data that drives it, and design systems that can adapt in real time. That means rethinking team structures, delivery cycles, testing strategies, and even how we measure success.

I presented this at a Data Visualization Meetup in DC (~70 people in attendance, 40min), and the feedback was overwhelmingly positive. In addition to discussing why someone might want to visualize data in AR/VR, I’d like to demo how this type of visualization works. I’ll bring a VR headset and also let people try it in AR mode on their phones.

Fireside Chat with Lior Hacohen, we will discuss compliance from a security perspective and the importance of early implementation.

Product matching i.e. matching the same product across retailers is key to driving revenue in our marketplace model — by identifying duplicate listings, we can prioritize retailers that best work for our business. Other industry use cases include catalog consolidation, recommendations and price comparison. However, it is a notoriously hard problem to solve - product data often varies in structure, terminology, and completeness across sources. In this talk, we'll share our prototype system that combines text and image embeddings, machine learning, and LLM-as-a-judge to improve match precision. Attendees will learn how this led to more matched products and incremental revenue.

Are you ready to build a futuristic solution capable of optimizing a repeatable business process? Are you a developer, data scientist, designer, tinkerer, or just a tech enthusiast ready to explore the next frontier of AI? At the DevFestDC 2025 Agentic AI Challenge , you can form a team and spend the day bringing your innovative solution to life! We'll dive deep into the emerging world of Agentic AI - intelligent systems that act, adapt, communicate, and solve complex real-world challenges autonomously. Join the best minds from startups, corporations, and academia as you tackle use cases in government, finance, travel, consumer, autonomous vehicles, manufacturing and other enterprise functions including IT.

Learn how to build Agentic AI applications with Google's Agent Development Kit (ADK).

Product matching i.e. matching the same product across retailers is key to driving revenue in our marketplace model — by identifying duplicate listings, we can prioritize retailers that best work for our business. However, it is a notoriously hard problem to solve - product data often varies in structure, terminology, and completeness across sources. In this talk, we'll share our prototype system that combines text and image embeddings, machine learning, and LLM-as-a-judge to improve match precision. Attendees will learn how this led to more matched products and incremental revenue.

Keras has long been loved for its ease of use. Beyond that, it now has new capabilities to power your generative AI workflows. In this talk, learn how to use KerasHub to load model checkpoints from HuggingFace Hub, perform function calling from your favorite LLM, and see the flexibility of working with PyTorch, JAX, and NNX. Now you can enjoy the familiarity and well designed APIs of Keras while working on your latest GenAI ideas.

Advanced Video Experiences: Next-gen video codecs and modern HDR formats in Android Multimedia

Advanced Video Experiences: Next-gen video codecs and modern HDR formats in Android Multimedia

Resilience & Reliability in current AI Adoption curve across enterprises.

APIs gave apps data. MCP (Model Context Protocol) gives apps intelligence. With MCP, developers can let AI models directly connect to their apps and tools, making them usable as part of a broader AI ecosystem.

In this session, we’ll demystify how MCP works and show how to make your app “AI-ready.” You’ll see how to:- Expose your app’s functions and workflows as MCP tools - Enable AI models like Gemini to call into your app for real-world actions- Build secure, context-rich bridges between your app and external AI systems - Demo: a Flutter app that acts as a travel assistant, where Gemini (via MCP) retrieves flight data and executes actions inside the app.

In the near future, apps won’t just talk to users: they’ll collaborate with AI agents. This session provides a clear roadmap and working example for making your apps first-class citizens in this new AI-driven ecosystem.

Learn how to build Agentic AI applications with Google's Agent Development Kit (ADK).

In the new AI-enabled coding environment, vibe-coding is both an enticing and often-misunderstood reality. Learn key insights through the startup phase in the AI-enabled coding world.

Welcome to a candid engineer-to-engineer discussion about today’s reliability challenges—and why AI is no longer optional, but foundational.Developers are expected to fail fast, recover faster, and now think smarter.Speed is non-negotiable, but what’s really reshaping the game? AI—no longer just another tool, but a strategic partner developing a solution or writing a code. Let’s be real: developers have to channel their inner startup founder now—test relentlessly, adapt on the fly, and figure out how to push boundaries inside complicated frameworks.

This Fireside chat isn’t about tech hype; we’re tackling the realities of how AI is revolutionizing how to build solutions that are reliable. We’ll confront the tough questions:
•  Are we building smarter teams, or just accelerating the pipeline?
•  What does real developer autonomy look like when AI is making decisions in real time?
•  Which new skills and mindsets are essential for developers to keep control in an AI-driven environment?

Let’s talk about how you lead it, not just survive it.

In an AI-first world, everyone can be a builder. But building is just the beginning. Shipping and running reliable applications that users love is what matters. This session will reveal a critical truth: AI's primary role is that of an amplifier. It magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones. Let's look at how your team can scale brilliance and eliminate bottlenecks, enabling builders everywhere

Picture two roller derby coaches at opposite ends of a bench, both holding tablets and both needing to see the same jam unfold in real time, with no backend team, no compromise on the live experience, and a third read-only tablet they can hand to whoever wanders over. That’s Bench Witch, and the underlying problem is everywhere: collaborative whiteboards, multi-user POS systems, shared dashboards, and any product where multiple clients need to edit shared state in real time.

Artificial intelligence is no longer confined to the cloud or screens - it is beginning to live inside the physical world. From smart wearables that monitor heart rate and detect anomalies in real time to industrial power tools that automatically adjust torque under changing load conditions, intelligence is shifting closer to where data is created and actions actually happen. This shift, often called edge computing, or physical AI, is redefining how we think about intelligence in everyday systems.

Kubernetes has changed how we manage cloud and on-prem environments. But some aspects of the node lifecycle have not fully embraced the Kubernetes API. This has meant context-switching between IaC tools and Kubernetes which adds unnecessary complexity. Karpenter aims to modernize node autoscaling and lifecycle management in Kubernetes: it auto-discovers cloud capabilities, launches the best hardware for the job, and powers it down when it's no longer needed. Come learn about Karpenter and see it in action!

It’s Monday morning and the Accounts Payable team gets a Slack message: “Why was invoice INV-2291 flagged, and how much exposure do we have with this vendor?” The data is in a legacy Purchase-to-Pay (P2P) system — dozens of tables, no data dictionary, weak foreign keys, and a schema that only one retiring analyst ever fully understood. Unfortunately, a generic chatbot would only give guesses or a junior analyst would spend half the day tracing data across joins nobody documented. I will show you how I built a P2P Context Engine to handle exactly that kind of question — end to end, with financial precision.

Prompt Engineer, so hot right now! Do you wonder why we need an entire specialization dedicated to communicating with a computer? Does it feel like "it just doesn't get you" when "chatting" with AI tools?

In this session, we'll cover the history of ways humans have communicated with computers. You'll see why "natural language" doesn't seem to translate as you expected. Finally, you'll learn options for creating an interface with AI that doesn't require a special title. Along the way, you'll learn to create trust and reduce hallucinations.

#AirClaw demonstrates how Apache Airflow and NVIDIA NemoClaw transform static pipelines into autonomous agents that reason, act, and escalate intelligently. Two live demos ( a NYC 311 overnight triage agent and an automated LLM model migration evaluation comparing GPT-4o against Llama-3.3-Nemotron-Super ) show the same three-layer architecture solving real production problems across two completely different domains.

The AppFunctions Jetpack framework allows apps to expose structured, callable capabilities directly to the OS and on-device AI agents. This marks a fundamental shift away from fragile screen scraping and rigid intents, enabling true agentic development where models natively discover and invoke your app's features.

We will demystify how AppFunctions work and contrast them with older approaches like App Actions and MCP. Through a focused code demo, you will see how to define typed, callable functions that allow agents to execute tasks straight from natural language prompts.

I vibe coded a consumer iOS app and shipped it solo, ~5,000 downloads, ~33% paid conversion,  zero ad spend. This talk is everything those numbers didn't teach me until after I shipped.

I'll walk through the real lessons from taking AI-assisted code into production: the layer AI silently skips (error reporting, and the analytics that actually predict retention and conversion), the AI-generated code that broke in ways I didn't see coming. Real dashboards, real crash logs, real numbers, including the app I'm still getting wrong right now.

What if an AI could read a novel and hand you an itinerary? That's the question behind My Storyland — a multi-agent Gemini system that extracts real-world locations from literary works and transforms them into personalized travel experiences.

In this talk, I'll walk through how I designed and built a collaborative agent architecture — Guide, Book Searcher, Location Extractor, and Composer — that doesn't just chain prompts, but assigns distinct roles, manages context handoffs, and synthesizes outputs into something genuinely useful. I'll cover how I evaluated agent performance using Langfuse, why human-in-the-loop checkpoints matter even when your agents "work,"  and how Pydantic-validated schemas keep your agent outputs structured and trustworthy instead of free-form and fragile.

Agents, recommenders, and reinforcement-learning loops all tune themselves at runtime — and most ship with no guarantee they'll settle rather than oscillate, drift, or quietly degrade until it surfaces in production. We're good at building systems that adapt, and bad at proving they'll behave.

Control engineers solved a version of this decades ago. Lyapunov stability theory proves a system converges to its target instead of assuming it will — the same principle that keeps a self-driving car centered in its lane. This session brings it to the systems we build now, and is honest about the catch: a proof is only as strong as the model beneath it. The math is rigorous; the assumptions are where the risk hides. That line — what the proof buys, and where it stops — is the talk.

If your multi-agent pipeline forces users to watch a loading spinner while it thinks, you have already lost them. But if your agent hallucinates a high-pressure gas line on a city street, you have a much bigger problem. Aligned with the "Build Edition" theme, this session strips away the "Hello, World" chatbot boilerplate to walk through the architecture of The Invisible City—a real-time, see-through AR visualization of underground municipal infrastructure. We will dissect how to pair Gemini’s multimodal vision capabilities with a Next.js and FastAPI stack to build event-driven agents that actually interact with the physical world.

A debate is raging in the XR community - which is the "correct" form factor for Intelligent Eyewear? Should we be focusing on the ambient experience from "AI Glasses" such as the models now appearing from Google, Samsung, Gentle Monster, and Warby Parker? Or should we have more of an immersive AR or VR experience, such as those from XREAL or the Apple VisionPro? Allen and Noble continue their many-year-long debate on this topic, last heard at DevFest DC in 2017. Which is the right technology?

Maps, search, navigation, real-world data — your users expect it all, and Mapbox delivers it in a single location platform built for Android developers. In this session, we'll start from zero and build up a fully featured location experience live: custom map styles, GeoJSON overlays, a working Search Box integration, and a turn-by-turn navigation flow with simulated routes and voice. We'll also cover what's new across the Mapbox ecosystem and show off the developer tools that help you move from idea to shipped feature, fast.

Everyone has a favourite framework. Nobody has tried all of them on the same problem.

You learned LangChain from a tutorial. Your colleague swears by ADK. Someone says AgentKit is the future. CrewAI keeps showing up in multi-agent demos. And your manager asks why you are writing code when Vertex AI Agent Builder exists. Each one solves a different problem and picking the wrong one means weeks fighting the framework instead of building your product.

In this session, we will explore how to design a secure tool-calling workflow using Gemini, Agent Development Kit, and Google Cloud. Attendees will learn how to move from a simple chatbot prototype to a production-ready agent architecture that can safely call enterprise APIs, return structured responses, and log actions for review.

In this talk, we will explore the architecture of local (stdio) and remote (HTTP) MCP servers within the Antigravity suite (2.0, IDE, and CLI). We’ll walk through how to build a custom MCP server from scratch, securely handle authentication via Application Default Credentials (ADC), and set up strict access control policies. Attendees will leave with a concrete understanding of how to transform an AI from a text-generator into an autonomous development partner capable of safely executing data-aware workflows.

What happens when your AI agent needs to join a queue of 14,000 people, wait 40 minutes, survive a process crash mid-wait, and spend $400 of real money while you’re asleep?

Most agent tutorials teach you how to prompt. This hands-on workshop teaches you how to build long running agents using Google Cloud and the Agent Development Kit (ADK 2), you will hands-on build a ticketing agent from scratch and then break it. We'll pull the plug mid-execution, simulate race conditions, enforce financial spending fences, and deploy a resumable agent architecture that runs without you in the room.

Your codebase was fine. Then you gave AI a taste. Join us for this honest, and slightly chaotic look at what happens when you let AI loose on real production code. I'll dig into what worked, what exploded, and what I would never admit to my tech leads.

I'll cover the full arc — from writing prompts that actually work, to understanding why agents are a completely different beast from chatbots, to the $0.60 mistake that taught me more than any tutorial ever did. Whether you're an AI-curious skeptic or someone who's already handed the keys to OpenClaw, you'll walk away with the habits that separate people who use AI like a professional from everyone else still using it like a tourist.

AI is moving fast—and policy is racing to catch up. This fireside chat explores the latest developments in AI policy, what developers should expect next, and why it matters.

AI is moving fast—and policy is racing to catch up. This fireside chat explores the latest developments in AI policy, what developers should expect next, and why it matters.

AI is moving fast—and policy is racing to catch up. This fireside chat explores the latest developments in AI policy, what developers should expect next, and why it matters.

Challenges, considerations, and capabilities in building agentic systems

Challenges, considerations, and capabilities in building agentic systems