February 24-26, 2027
Montreal, Canada
Artificial Intelligence Conference
The list of presentations is subject to change without notice.
Artificial Intelligence
Working with natural language has always been complex. LLMs have made it more accessible, but production applications still face significant challenges, especially when dealing with multiple languages. This talk covers common pitfalls in multilingual audio and text workflows, along with practical approaches for selecting, evaluating, and combining speech-to-text and language models.
Artificial Intelligence
In June 2026, OpenCV released version 5.0, its first major release since 2018. This talk uses live demos to explore its core features and newest additions. From Hough transforms and image stitching to object detection and facial recognition, attendees will see what OpenCV can do and how to apply it in their own projects.
Artificial Intelligence
As autonomous agents replace manual web browsing, e-commerce requires standardized, machine-readable interfaces. We will go over an end-to-end framework implementing the Universal Commerce Protocol (UCP) for agentic commerce. Using the Model Context Protocol (MCP), our system connects LLM agents directly to Shopify’s global catalog, enabling structured product discovery, live cart management, and autonomous checkout orchestration.
Artificial Intelligence
AI-enabled in vivo systems can sense physiology, infer state, and trigger physical action. I examine poisoned inputs, compromised updates, sensor/model drift, and unsafe autonomous decisions. I show how independent monitors detect loss of trust, how bounded actuation and external override contain it, and how known-good recovery and re-attestation restore control. Attendees leave with a reusable fail-safe pattern for safety-critical edge AI.
Artificial Intelligence
How much code do you still truly read? Do you check what green tests actually prove? Everything passed review without a problem, but the outage it caused is still yours. I will teach you my hard-won principles for getting this process back under control: verifying without asking, checking assumptions before building on them, ranking changes by irreversibility, actually enforcing rules. You wanted to make things predictable. Let's reclaim that.
Artificial Intelligence
I built a self-driving SDLC that specs, codes, tests, reviews and ships - and battle-tested Temporal, Restate, Inngest, LangGraph and DAG schedulers to get there. Durable execution wants a world that never changes; an LLM invents a new answer every call. Keep them apart - a thin controller picking the next step, model calls and file writes outside it - or a crash re-runs your expensive calls and pays twice. Here's the scorecard and the scars.
Artificial Intelligence
Diffusion models excel at short clips but suffer severe drift across extended sequences. We examine agentic video frameworks that solve this orchestration bottleneck. By decoupling production into hierarchical agents for narrative planning, identity-locked keyframing, audio-driven pacing, and code-based assembly, these systems preserve continuity, offering a scalable blueprint for autonomous, long-format video generation.
Artificial Intelligence
From ChatGPT text boxes to Claude Code and Codex, LLMs evolved from basic chat to tab completion to active digital workers. The secret isn't bigger models: it's the harness. We will build a complete harness live from scratch in Python, demonstrating how it bridges next-token prediction with managed state, action execution, and error handling. Plus, we'll peek at how these execution loops generate training data for future models.
Artificial Intelligence
As coding agents become integral to daily software development, whether you're a seasoned engineer or a new builder looking to launch your first app, how you build and what you build is fundamentally changing. Learn how to work through the people-pleasing bias of agents, ensure they can verify their work correctly, and how to use modern web APIs and patterns to reduce technical debt!
Artificial Intelligence
Remote Model Context Protocol (MCP) servers expose tools, but did you know they can return UI, not just text? Using the MCP Apps extension, an MCP app delivers interactive experiences inside hosts like Claude.ai and ChatGPT.
We'll live code one from scratch, tunnel it to a public URL, register it in both hosts, then lock down tool calling with Pomerium (open core, Apache 2.0) so you control who gets to use your app and its tools.
We'll live code one from scratch, tunnel it to a public URL, register it in both hosts, then lock down tool calling with Pomerium (open core, Apache 2.0) so you control who gets to use your app and its tools.
Artificial Intelligence
Rolling out an AI coding agent takes more than installing a CLI. I'll show how far OpenCode's remote config, plugins, and skills go in an enterprise setup: one-command onboarding, layered configs, and a shared plugin monorepo. I'll explain how we set agent permission defaults, manage model access, and track cost and usage through telemetry. You'll leave knowing exactly what OpenCode provides and what your team needs to build.
Artificial Intelligence
AI-powered tools can detect accessibility issues faster than ever, but they cannot replace human judgment. This presentation examines the strengths and limitations of AI for WCAG testing, highlights common pitfalls, and demonstrates how AI can support accessibility professionals, QA teams, and developers in delivering more inclusive digital experiences.
Artificial Intelligence
An AI Agent that immediately fixes the tests can hide a real product defect. In this session, I'll show you a Playwright suite where an Agent reads a failed GHA run, classifies locator drift versus a real regression, and refuses to soften assertions. You will see a deterministic hook reject a weakened expect, and how the Agent decides to fill a Jira bug instead of healing the test.
Artificial Intelligence
Running an autonomous agent like OpenClaw without hardening access to it is a bad idea. This talk covers securing OpenClaw ingress: OIDC login, short-lived SSH certificates, per-connection policy, and live session revocation, using Pomerium, an open core (Apache 2.0) identity-aware proxy built on Envoy, a CNCF project. Includes a PR I contributed upstream (trusted-proxy auth mode) and a live demo killing an active session mid-task.
Artificial Intelligence
LLM agents often feel random, forgetful, or fragile, but in most cases the problem is the context, not the model. Prompts, tool output, memories, and failed hypotheses all compete for attention. This talk shows how clean context makes agents more reliable, cheaper, and easier to trust, with practical patterns like forking, filtered CLI output, and careful MCP usage to keep agents focused and out of the token landfill.
Artificial Intelligence
Building AI agents on top of REST APIs sounds straightforward: turn each endpoint into an MCP tool. But as APIs grow, tool definitions and responses crowd the model's context. MCP as Code offers another way: let agents discover APIs, write code to call them, and filter results before they reach the model. This case study explores how semantic search and sandboxed execution make that work, and where the approach helps or adds complexity.
Artificial Intelligence
Modern AI coding assistants like GitHub Copilot and Cursor promise easy test automation; just prompt the assistant to write a test, and watch it work. I found the day-to-day experience much different. In this talk, I will show which tasks AI is suitable for, how to collect relevant context for each prompt, and how to guide the AI to code better end-to-end tests.
Artificial Intelligence
Almost everything has changed since the mainstream adoption of generative AI. Every decision-making process must now factor in AI, from costs, usage, and standardization to risks and hiring. If you hold a technical leadership role, you inevitably had to adapt. I will share my observations and recommendations, but I am open to discussion.
Artificial Intelligence
LLMs coupled with coding agents can be very effective and are designed to be goal seeking. One of the ways to unlock this capability is to provide a computing environment to LLMs. There are many commercial products that offer sandboxes like E2B, Daytona, Docker Sandboxes etc. In this talk we will peel back the covers of how these work. We will see how to stage files for the sandbox, how to securely hoist secrets and limit network access.
Artificial Intelligence
LLM assistant running inside your app breaks a core assumption of e2e testing: the same input can produce different tool calls. I'll show you how I tested an in-app AI assistant using one Gherkin contract and three runners: direct handler calls, deterministic tests using Playwright, and agent-driven browser testing.
Artificial Intelligence
Modern web development provides an alternative to cloud-based round trips for AI usage and enables user to co-browse your site with Agents. Learn how you can take advantage of in-browser, on-device models for prompting and on-the-fly language translation and build out agentic experiences with WebMCP to enhance an e-commerce site, culminating in creating an on-site agent that works entirely offline!
Artificial Intelligence
As AI adoption grows, it increasingly encounters Java-based business logic and data. Java is everywhere AI needs to be. This session presents an end-to-end overview of AI and the Java Platform from source-code generation, tooling, inferencing, native GPU compute from Java, and everything in between. We will explore Java innovation projects like Panama, Babylon, and Valhalla, and how they benefit the Java ecosystem and AI workloads.
Artificial Intelligence
Foundation models did not stop at text. Pretrained on billions of points, time series models now forecast series they have never seen, with no training at all. We put one in production and retired the models we trained per metric. You will learn how zero-shot forecasting works, when it beats training on your own data, and what it costs you.
Artificial Intelligence
Agent harnesses are easy to run on your machine. Giving them a secure place to work on their own is harder. Using Modal Sandboxes, this talk walks through the move from local agents to isolated, sandboxes with controlled access to code, tools, secrets, and networks. You will learn the building blocks for running harnesses as persistent engineering collaborators instead of as personal assistants.
Artificial Intelligence
This session explores how AI agents move from demos into real-world environments. Drawing on technology first used for holographic communication with the ISS and later deployed in airports, healthcare, and training, I’ll share lessons on AI Presence, context-aware systems, edge vs. cloud decisions, unreliable networks, privacy, and post-quantum security. Attendees will learn practical approaches for building resilient, aware, secure AI.
Artificial Intelligence
Building applications with AI and large language models (LLMs) adds a new layer of variability that makes testing difficult. Since interactions can be subtly different each time but still correct, traditional testing tools often fail. In this talk, we'll share our journey using the open-source tools DeepEval and MLflow to evaluate multi-turn conversations, tune prompts, and ensure our AI applications are reliable so that we can sleep at night.
Artificial Intelligence
Want to add AI capabilities to your Java applications but don’t know where to start? This beginner-friendly session introduces LangChain4j, a framework for integrating large language models into Java applications. Learn how to connect to a model, build a conversational assistant, and use your own documents to ground its responses with retrieval-augmented generation (RAG).
Artificial Intelligence
Developers already ship code with agentic AI and MCP servers without governance. Blocking them kills velocity; ignoring them creates risk. This session shows how to build a centralized MCP broker with Amazon Bedrock AgentCore Gateway and Runtime, enforce identity-based permissions on agent requests, distribute a curated skills catalog with reusable blueprints, and embed security agents into the AI SDLC so you can govern AI without slowing it down
Artificial Intelligence
You've likely heard how rising AI bills have made teams start thinking about running their own open-weight models. The good news? You can achieve near-frontier capabilities, but the hard part is figuring out the optimizations that are right for your hardware, from Mac's to NVIDIA GPU's. I'll walk you through what works best, from quantization to speculative decording, through live demos that balance throughput, latency, and accuracy.
Artificial Intelligence
The largest, smartest models are also the most expensive to run; quantization is one of the most accessible ways to cut that cost. This talk introduces model quantization technically, with a live pipeline demo: LLM Compressor handles quantization and sparsity, GuideLLM benchmarks latency and throughput against baseline, and lm-eval scores accuracy. Attendees leave with open source tools to compress their own models without losing performance.
Artificial Intelligence
Human-in-the-loop might be the easy answer, but a better model is where people set direction, agents execute, and review at key checkpoints. The real skill is deciding where a human has to own the call. This session works through that using a personal night-sky viewing planner application and weighing where AI should or shouldn't take the wheel. Attendees will leave with skills to decide where AI should drive and where it should ride along.
Artificial Intelligence
L'IA agentique passe du MVP à la PROD plus vite que la plupart des équipes sécu n'adaptent leurs modèles de menace.
Des agents autonomes qui appellent des outils, enchaînent des LLMs et communiquent entre eux via des protocoles comme MCP et A2A introduisent des surfaces d'attaque que l'OWASP Top 10 classique n'a jamais couvertes : le prompt injection comme vecteur de mouvement latéral, le tool-poisoning, l'exploitation de la confiance A2A
Des agents autonomes qui appellent des outils, enchaînent des LLMs et communiquent entre eux via des protocoles comme MCP et A2A introduisent des surfaces d'attaque que l'OWASP Top 10 classique n'a jamais couvertes : le prompt injection comme vecteur de mouvement latéral, le tool-poisoning, l'exploitation de la confiance A2A
Artificial Intelligence
Open Mercato is an MIT-licensed CRM/ERP framework whose code is largely written by AI agents. As a contributor to it we can together do a tour from the outside in: specs committed beside the code, shared agent skills for pull requests, review and CI stabilization, conventions agents cannot drift from, and where humans stay in the loop. Everything is public, so we look at what holds, what breaks, and what transfers to any stack.
Artificial Intelligence
Step into JARVIS voice-first mission control for Gemini’s multi-agent systems. Explore Gemini Live, ADK 2.0, A2A, and MCP powering real-time agent teams: heartbeat loops, voice intents spawning long-horizon tasks (tickets, research). Unlock the control-panel pattern for steering agent swarms, with production tactics for bidirectional voice, tool use, and governance. AI orchestration as it was meant to be.
Artificial Intelligence
Tired of cloud API latency and rate limits? Learn how to run, quantize, and serve open-weight models directly on your workstation using modern runtimes, optimal precision formats, and practical hardware sizing. We'll cover:
* How to choose the right quantization method for your target memory budget
* How to select and configure the right inference engine
* Practical rules of thumb for memory bandwidth, VRAM sizing, and context window
* How to choose the right quantization method for your target memory budget
* How to select and configure the right inference engine
* Practical rules of thumb for memory bandwidth, VRAM sizing, and context window
Artificial Intelligence
Microsoft Agent Framework has plenty of “hello world” demos, but the real leverage often comes from the features that don’t show up in most talks. In this session, we’ll skip the basics and focus on the less common topics that make these kinds of solutions more usable, maintainable, and production-ready.
By the end of the session, you'll have a short list of techniques you can apply immediately to level up your MAF game.
By the end of the session, you'll have a short list of techniques you can apply immediately to level up your MAF game.
Artificial Intelligence
When an AI agents goes wrong, it doesn't fail in one log line. It's across a chain of model decisions, retrieval, retries, and tool calls that jump between services. So, how do we tell what happened with each request, and perhaps why the response wasn't what we wanted? This session is a demo of tracing AI agents using OpenTelemetry and MLflow to understand requests from user -> agent -> tools -> and back, to better understand agent behavior.
Artificial Intelligence
Everyone seems to be building agentic AI systems with multi-step workflows, autonomous decision-making, and agentic processes. So what does that actually look like in practice? In this session, we will build a real agentic Java application, walking through how it works and when agents need external help. Come see what it actually takes to move from a single tool call to a genuine agentic workflow for your own Java projects.
Artificial Intelligence
Le téléphone de vos utilisateurs embarque déjà un modèle performant. Regardons ensemble ce qu'on peut bâtir aujourd'hui avec les Foundation Models sur iOS et Gemini Nano sur Android : résumés, données structurées et tool calls, le tout hors ligne, privé et surtout, sans aucun coût de token.
Artificial Intelligence
Tanné des faux positifs et des failles manquées par vos outils de sécurité? Vous avez essayé les LLM, mais sans la magie promise?
Cette présentation propose un pipeline complet, pas juste des invites (prompts), pour exploiter efficacement les LLM en sécurité applicative. Réduisez vos faux positifs et repérez les failles ayant un réel impact.
Cette présentation propose un pipeline complet, pas juste des invites (prompts), pour exploiter efficacement les LLM en sécurité applicative. Réduisez vos faux positifs et repérez les failles ayant un réel impact.
Artificial Intelligence
We will show what options exist to run thousands of agents, focusing on Open Source and Kubernetes options. We will cover their full lifecycle (provision, execute, observe, and tear down), while keeping workloads isolated from the rest of the platform. Agent sandboxes enforce resource limits, reduce blast radius, and give operators a consistent place to apply policy.
Artificial Intelligence
AI enables us to generate code faster than ever – but delivery pipelines often struggle to keep up. LLMs can produce brilliant solutions as much as spectacular noise, flooding teams with more changes than they can reasonably absorb. The result: CI pipelines become the bottleneck, wasting cycles on redundant and often pointless work, returning feedback long after it’s useful.
This talk explores how to reshape CI/CD for AI-accelerated development.
This talk explores how to reshape CI/CD for AI-accelerated development.
Artificial Intelligence
AI coding agents fail on ambiguity, not syntax. BDD solved that years ago with executable specifications in business language. This talk maps a decade of BDD practice onto spec-driven development with agents: scenarios an agent implements against, used as the acceptance gate, and how to catch false green when the agent writes its own tests. Examples come from real projects and from building an open-source ERP. The workflow fits any language.
Artificial Intelligence
IT tickets pile up because rule-based automation can't gather missing info or use judgment. This talk shows how AI agents embedded in a ticketing system triage requests, ask questions, and resolve common cases end-to-end, cutting response time and backlog. We'll cover specialist vs. general agents, knowledge-base grounding, guardrails against prompt injection, evaluating non-deterministic agents, and tracing conversations for full observability.
Artificial Intelligence
AI coding assistants have evolved quickly, and they are not slowing down. New models, tools, MCP servers, and techniques appear constantly, making it easy to feel like you are always behind.
Instead of chasing every new update, this session focuses on how to build one that can evolve.
I will cover:
- Understanding the context window
- Skills and MCP servers
- My agentic workflow
- Managing costs
- Systematic approach to evolving yours
Instead of chasing every new update, this session focuses on how to build one that can evolve.
I will cover:
- Understanding the context window
- Skills and MCP servers
- My agentic workflow
- Managing costs
- Systematic approach to evolving yours
Artificial Intelligence
Hugging Face is far more than a model hub—it’s a complete cloud solution for all your AI needs. Discover how Buckets and Datasets manage data, Jobs power compute and training, and Inference Providers deliver instant production serving. Tied together with Spaces for UI deployment, see how these services work seamlessly as a unified platform to build, train, and scale end-to-end AI workflows.
Artificial Intelligence
Your agent finally remembers things. Congratulations, it is now a data leak with a personality. The whole room gets a link to one agent and your job is to ruin it: feed it lies, leak someone else's context, make it hold a grudge. We watch its memory fill up, then fix each failure with boring, fast tools: streams, structured records, vector search and expiry. You leave knowing what an agent must forget, and why that is where production breaks.
Artificial Intelligence
Even with robust CI/CD, production rollouts can hit unexpected snags. While in Kubernetes, Argo Rollouts excels at Progressive Delivery, what if we could go a step further?
This session explores how to elevate your release process by integrating Agentic AI with Argo Rollouts canary deployments. We'll demonstrate how an intelligent agent can automatically analyze and propose a fix for a rollout failure, pinpointing the root cause.
This session explores how to elevate your release process by integrating Agentic AI with Argo Rollouts canary deployments. We'll demonstrate how an intelligent agent can automatically analyze and propose a fix for a rollout failure, pinpointing the root cause.
Artificial Intelligence
Your website already has an API for AI agents. It's called the DOM, and agents are awful at using it. Right now an agent on your site screenshots the page, guesses which button means "checkout", and hopes. In this talk I take a real site and, live, expose its search, forms and checkout as tools with WebMCP, the browser standard Google and Microsoft are pushing through Chrome. You leave able to make your own site agent-ready in an afternoon.
Artificial Intelligence
Most AI tools make one developer faster in a private session. This talk explores what changes when AI becomes a teammate in Slack. Drawing on our experience building Luma at Lightspeed, I’ll cover shared context, tools, memory, safety and auditability. Beyond the technical implementation, we will explore how public AI collaboration helps teams reuse investigations, learn from each other, discover new workflows, and adopt AI organically.
Artificial Intelligence
Your API was designed for a caller that reads the docs. An LLM does not. It sees an empty array and invents a reason for it. This is a talk about designing responses for a non-human caller: errors that carry a recovery plan, empty results that say whether nothing exists or you simply cannot see it, why 404 can be the wrong status code, and why one polymorphic mega-tool loses to several narrow ones. We'll look at real code from a live MCP server.
Artificial Intelligence
AI is powerful and eager to please. It will go anywhere to answer your question, including places it's not supposed to. Your job is to control what it can see, and what it can't. Join us and learn about the complicated world of authentication and authorization for AI. We'll dig deep into auth types, roles, grants and audit trails. From the high-level to annoying details, learn to build agents that are powerful, but respectful of their boundaries.
Artificial Intelligence
Your job is simple: make the ultimate AI oracle that can answer any question by querying every data source in your org. But your users don't even want it - they want to connect their own AI. Enter ontologies: detailed maps of your data and what it means. Join us and learn to shed light on dark schemas, surface better insights and make a truly great AI oracle. You'll make something that's not only useful, but something they'll pay for as well.
Artificial Intelligence
AI-native engineering changes more than how developers write code. This talk explores the operating model behind it: shared intelligence, context as code, reusable skills, model routing, evals, traceability, and earned autonomy. We will look at how planning, governance, and human judgment fit together as agents take on more of the SDLC. You will leave with a framework for designing the transition.
Artificial Intelligence
Vector search is moving from specialized systems into mainstream databases such as PostgreSQL, MySQL, MongoDB, and Elasticsearch. This talk explains how embeddings are created, stored, and indexed; when to combine vector and traditional search; and how to measure speed and result quality. Attendees will gain practical guidance for choosing an architecture and building effective AI-powered search applications.
Artificial Intelligence
AI agents aren't just writing code anymore. They're executing it. I gave agents access to real cloud infrastructure and turned off the guardrails. This talk shares five war stories of what went wrong, a framework for thinking about when agents are safe to use, and the benchmark we're still missing. You'll leave with concrete safeguards for your own agent setup.
Artificial Intelligence
AI is becoming capable of using more than words. This session explores how multimodal systems can combine voice, visual, behavioral, and environmental cues to understand context and adapt their responses. I’ll discuss affect-aware AI, privacy boundaries, bias, human trust, and lessons from building Aexa Aware. Attendees will learn how to design AI that is more context-sensitive and human-centered without pretending machines actually feel emotion.
Artificial Intelligence
Imagine anyone in the business asking questions of the database like they're talking to a person. No SQL. No ticket. No waiting for a developer.
We built it: an MCP with read only DB access that hid PII. Every client who has tried it is hooked.
But AI also finds fool's gold. I'll show how to build one, how to stop confident wrong answers (well, most of them), and how to turn the best questions into trusted reports.
We built it: an MCP with read only DB access that hid PII. Every client who has tried it is hooked.
But AI also finds fool's gold. I'll show how to build one, how to stop confident wrong answers (well, most of them), and how to turn the best questions into trusted reports.
Artificial Intelligence
Passing evals and rising velocity feel like proof that an AI feature works, but both can mislead. This talk explains 4 ways offline evals deceive teams, including an LLM judge that rewards confident wrong answers, then fixes one live with a small rubric change. Attendees learn what to measure instead: traces, production observability, probe-based evaluation, error budgets and quality indicators beside every velocity number.
© 2010-2026 ConFoo. All rights reserved. Code of Conduct
Cookie Consent
To provide the best experiences, the usage of cookies are necessary. Some are essential for the proper functioning of the site, while others help us better cater to your interests.