SPEAKER NOTES - Title Slide (2 minutes) Welcome and Introduction: - Introduce yourself warmly - "And who here has experimented with AI assistants like ChatGPT or Claude in their scientific work?" Set the stage: - "Today I want to show you how these two powerful technologies - OME-NGFF and AI - are converging to transform how we work with bioimage data" Timing: Keep this brief - 2 minutes max. Energy should be high!

SPEAKER NOTES - The Challenge (0.5 minutes) Make it visceral: - "Let me start with a situation you may have experienced..." - "The postdoc is excited - finally got those microscopy results" - "50 beautiful images, now you just need the quantification..."

SPEAKER NOTES - The Challenge (1.5 minutes) The old way: - "So they spend the next THREE WEEKS wrestling with analysis" - "Writing scripts, debugging when Image Number 23 has different dimensions" - "Re-running everything because they forgot to save parameters" - "This is the reality we regularly experience" The new way: - "Now imagine instead - they just DESCRIBE what they need" - "Natural language, like talking to a collaborator" - "The AI agent provides transformative technical assistance" - "Hours instead of weeks" Impact: - "That's three weeks your postdoc gets back for actual science" - "Three weeks closer to publication" - "And it's fully reproducible - no 'what parameters were used?' six months later" Question: - "How many postdoc-weeks would YOUR lab save?" - Let that sink in before moving on Timing: 1.5 minutes - this is your hook!

SPEAKER NOTES - Agenda (1 minute) Roadmap: - "We have 25 minutes together, so let's make them count" - Point to each section as you describe it - "First, I'll show you WHY OME-NGFF and AI are a perfect match" - "While we ARE admittedly only in the beginning stages of SCIENTIFIC AI, we will dive into a REAL tool you can use today - the ngff-zarr MCP server" - "Finally, we will look at concrete steps YOU can take NOW to get started" Timing: 1 minute - keep it moving!

SPEAKER NOTES - Part 1 Transition (30 seconds) Transition: - "Let's start with the big picture - what's happening in AI right now" - Energy shift: move from logistics to inspiration - "Let's examine why OME-NGFF is ideal for agentic scientific AI" Timing: Quick transition - 30 seconds

SPEAKER NOTES - AI Revolution (1.5 minutes) Context setting: - "We're in the middle of a profound shift in how we interact with computers" - "Most folks have used chatbots to search for information or generate narritive text" - "But here's what's NEW - we're moving beyond chatbots to AGENTS" Key point - AGENTIC AI: - Emphasize the difference: "Not just answering questions, but actually DOING work" - "Planning multiple steps, using tools, making decisions" The challenge: - "But here's the problem - how does AI -- text-based large language models -- actually WORK with YOUR imaging DATA?" - "That's where OME-NGFF comes in..." - Pause for effect before moving on Timing: 1.5 minutes - this is foundational!

SPEAKER NOTES - Agentic AI Definition (1.5 minutes) Break it down clearly: - "Let me make this concrete with an example..." - Walk through each element: * Context: "The AI knows about YOUR specific microscope, YOUR file formats" * Tools: "It can actually run conversion software, optimization scripts" * Reasoning: "It plans: first validate, then convert, then optimize" * Execution: "And it DOES it - not just tells you how" Engage with questions: - "Think about your last batch conversion task - how long did it take?" - "What if you could just DESCRIBE what you need and have it happen?" Make it tangible: - "This isn't science fiction - I'm going to show you how this works in 10 minutes" Timing: 1.5 minutes - enthusiasm is key here!

SPEAKER NOTES - OME-NGFF + AI Part 1: Cloud Architecture (1 minute) Connect to AI capabilities: - "Now here's where it gets interesting - OME-NGFF was DESIGNED for exactly this kind of intelligent access" Cloud-ready chunked storage: - "The AI doesn't need to download terabytes - it can grab just the chunks it needs" Hierarchical structure: - "Like reading specific chapters of a book instead of the whole library" Web-friendly, AI-tool-calling friendly interface: - "AI can request just the metadata it needs - dimensions, spacing, coordinate systems" - "Then fetch only the specific data regions required for analysis" - "Works seamlessly whether your data is local or in the cloud" - "The same HTTP-based interface works everywhere - perfect for AI tool integration" Timing: 1 minute - keep pace brisk

SPEAKER NOTES - OME-NGFF + AI Part 2: Metadata (45 seconds) Rich metadata advantage: - "But storage format isn't enough - the AI needs to UNDERSTAND your data" - "OME-NGFF embeds all the spatial metadata, calibration, channel information" Standards matter: - "Because it's standards-compliant, the AI can make intelligent assumptions" - "It knows what 'microns' means, what coordinate systems to use" Ecosystem: - "And it can chain together multiple tools that all speak OME-NGFF" Timing: 45 seconds - building momentum

SPEAKER NOTES - OME-NGFF + AI Part 3: Open Standards (45 seconds) Open science connection: - "And here's what I love most - this is TRULY open" - "Your data isn't trapped in some proprietary format" - "The AI tools we build today will work with your data in 10 years" Emphasize community: - "You're JOINING a COMMUNITY, not buying into a vendor" Timing: 45 seconds

SPEAKER NOTES - The Combination (1 minute) Bring it together: - "So let's put this together - what does this MEAN for your research?" - Point to the diagram: "Data format + AI capability = transformation" Make it real: - "Imagine: 'Convert these 1000 microscopy images, optimize for cloud storage, and validate the results'" - "Just describe it, and it happens" - "No bash scripts, no parameter tuning, no debugging at 2am" Each bullet: - Petabyte: "Datasets that would take weeks become lunch-break tasks" - Reproducible: "Everything documented, every parameter logged automatically" - Optimization: "The AI tries different approaches and finds what works best" - Batch: "Your weekend work becomes 10 minutes on Monday morning" Pause: - "This is the vision." - "We are not there yet." - "But we ARE on our awy there! Now let me show you how this ACTUALLY works..." Timing: 1 minute - this is the payoff slide!

SPEAKER NOTES - Part 2 Transition (30 seconds) Energy shift: - "Okay, enough theory - let's get practical" - "I'm going to introduce you to a free and open source tool available TODAY" - Check time - should be around 8-9 minutes in Timing: 30 seconds - quick transition

SPEAKER NOTES - MCP Introduction (1.5 minutes) What is MCP: - "The Model Context Protocol (MCP) is like USB for AI - a universal connector" - "Developed by Anthropic, but it's open and vendor-neutral" - Point to diagram: "Your AI assistant connects to scientific tools through this protocol" Why this matters: - "You don't have to choose between AI platforms" - "Works with Claude, with Cursor, with GitHub Copilot, with dozens of existing agent platforms and with future AI systems" - "The same MCP server works with all of them" Easy integration: - "And here's the beautiful part - YOU don't have to code the integration" - "The MCP server does all the heavy lifting" Question: - "Anyone here familiar with MCP already?" - If yes: "Great! You know where this is going" - If no: "You're about to see how simple it is" Timing: 1.5 minutes - this is new to most people

SPEAKER NOTES - ngff-zarr Introduction (1 minute) Introduce the tool: - "This is ngff-zarr - our contribution to the OME-NGFF ecosystem" - "It's a Python toolkit we've been developing at fideus labs" - "And now it has an MCP server that lets AI systems use it" What this means: - "You can talk to your AI assistant: 'Convert this image to OME-NGFF'" - "And behind the scenes, the MCP server translates that to ngff-zarr commands" - "Then executes them and reports back results" Emphasize: - "Completely open source - MIT licensed" - "Link is in the slides" Timing: 1 minute

SPEAKER NOTES - Capabilities (1.5 minutes) Go through each capability with examples: Convert: - "Supports all the formats you're probably using - TIFF, NRRD, Nifti, and more" - "AI can figure out the right conversion settings for your data" Validate: - "Check if your OME-Zarr files are spec-compliant" - "Catch issues before you publish or share data" Optimize: - "This is powerful - AI can test different compression and chunking strategies" - "Find the sweet spot for YOUR access patterns" Inspect: - "Get detailed information about multiscale structure" - "Understand what's actually in your data" Batch process: - "Generate bespoke batch processing scripts for large collections of images with reproducible settings" Timing: 1.5 minutes - let each capability sink in

SPEAKER NOTES - Before AI Example (1.5 minutes) Bring back the scenario: - "Remember our postdoc? Let's walk through their typical workflow..." - Make it painful but relatable Step by step pain: 1. Format conversion: - "First hurdle - get images into format the analysis tools can read" - "Proprietary microscope format doesn't work with open ecosystem tools" - "Spend half a day on format conversion alone" 2. Segmentation: - "Then the real work - write segmentation code" - "Parameters that work for Image Number 1 fail on Image Number 15" - "Back to parameter tuning, over and over" 3. Measurements: - "Finally getting data out - but wait, export format issues" - "CSV? JSON? How to link measurements to metadata?" 4. Statistics: - "Now wrangle everything into R or Python for stats" - "More data format conversion, more potential for errors" 5. Figures: - "Generate plots, realize you need different groupings" - "Re-run everything from step 3" 6. Documentation: - "And did you write down all those parameters?" - "Good luck reproducing this in 6 months" The reality: - "This is 2-3 weeks of a postdoc's time" - "And that's if everything goes relatively smoothly" - "We've all been there" Timing: 1.5 minutes - make them FEEL the pain

SPEAKER NOTES - After AI Example (1.5 minutes) The contrast: - "Now watch what happens with our agentic AI future..." - Read the natural language request slowly AI does the work: - "The agent helps us:" - "Analyze a sample to understand your data" - "Determine optimal settings based on your constraints" - "Generates AND RUNS the conversion" - "Tracks progress, handles errors" - "Gives you a summary when done" The magic moment: - "From hours or days to MINUTES" - "From error-prone to reproducible" - "From tedious to... well, you just describe what you want" Reality check: - "Now, this isn't magic - the AI CAN still make mistakes" - "But it's MUCH faster to review and correct than to do it all from scratch" - "And the agent can document everything as you work with the agent" Timing: 1.5 minutes - this is the "wow" moment

SPEAKER NOTES - In Practice Intro (15 seconds) Quick transition: - "Now let me show you what this looks like in practice" - "Three quick examples of natural language to results" Timing: 15 seconds - just a bridge

SPEAKER NOTES - Convert Example (45 seconds) Show the simplicity: - "Look at this - natural language, plain English" - "No command-line flags, no config files" - Point to the chat message: "Just describe what you want" Demo reference: - Click the link - "We see the agent's reasoning process" What happens: - "The AI agent takes over from here" - "Analyzes the file - dimensions, data type, characteristics" - "Picks appropriate parameters automatically" - "Runs the conversion and tests different codecs" - "Tells you which one worked best and why" Timing: 45 seconds

SPEAKER NOTES - Examine Example (45 seconds) Beyond conversion: - "It's not just for conversion - also for exploration" - "Ask questions about your data in plain language" Demo reference: - Click the link What you get: - "The AI inspects the multiscale pyramid structure" - "Reports all the spatial metadata - spacing, units, coordinates" - "Analyzes how the data is chunked and compressed for efficient handling of petabyte datasets" - "In the future, it could even suggest what you might want to do next" Use case: - "Great for validating datasets before sharing" - "Or understanding data someone shared with you" Timing: 45 seconds

SPEAKER NOTES - Batch Script Example (45 seconds) Scaling up: - "And here's where it gets really powerful" - "Once you've tested on one file, scale to many" Demo reference: - Click the link What it generates: - "It produces a complete, runnable Python script" - "Includes proper error handling - won't crash on bad files" - "Progress output so you know where you are" - "Uses the optimal settings it learned from your test conversion" Reproducibility: - "Now you have a script you can run again" - "Share with colleagues, include in your methods" - "Fully documented, reproducible workflow" Timing: 45 seconds

SPEAKER NOTES - Intelligent Conversion (45 seconds) What makes it intelligent: - "So what makes this 'intelligent' and not just scripted?" - "The AI adapts to YOUR specific data" Examples: - "Looks at dimensions, bit depth, noise patterns" - "Chooses blosc for speed or zstd for compression based on your needs" - "Generates clean multiscale pyramids automatically" - "Optimizes for local disk vs cloud access patterns" Bottom line: - "It makes informed decisions, not just defaults" Timing: 45 seconds

SPEAKER NOTES - Analysis and Reporting (45 seconds) Beyond conversion: - "Also handles inspection, validation, planning" Key features: - "Describes your multiscale structure in detail" - "Generates scripts for reproducibility" - "We are TRANSFERRING some of the BURDEN of REPRODUCIBILITY from a busy postdoc to the AI agent" - "Validates outputs before you share" Quick check: - Check time - should be around 18-19 minutes Timing: 45 seconds

SPEAKER NOTES - Future Vision (1 minute) Look ahead: - "This is just the beginning - where is this going?" Today vs tomorrow: - "Right now, the helps with format conversion" - "But we can have a future where you'll have advanced assistance through an entire analysis workflow" - "'Take these images, segment the cells, quantify fluorescence, compare to controls, generate publication figures'" Examples: - Multi-step: "Chain together preprocessing, analysis, visualization" - Optimization: "AI tries different parameters, finds best results" - Discovery: "And, the agent may even notice patterns you might have missed" Your role shift: - "You become the scientist who asks questions" - "Not the programmer who debugs bash scripts" - "Focus on WHAT you want to know, not HOW to compute it" Timing: 1 minute - inspire them!

SPEAKER NOTES - Part 3 Transition (30 seconds) Final section: - "We're in the home stretch - let's talk about YOUR next steps" - Check time - should be around 19-20 minutes - "How do you actually GET STARTED with this?" Timing: 30 seconds

SPEAKER NOTES - Immediate Actions (45 seconds) This week: - "Okay, concrete actions you can take THIS WEEK" Explore: - "Tonight in your hotel room - browse the docs" - "Look at the examples, see if they match your use cases" Try it: - "Tomorrow - take ONE image and convert it" - "Just pip install ngff-zarr and try the command line interface" - "5 minutes to get started" Evaluate: - "Think about YOUR workflow - where would this help?" - "What's your biggest pain point with image formats?" - "Could OME-NGFF solve it?" Timing: 45 seconds

SPEAKER NOTES - Short-term Goals (45 seconds) Next month or two: - "Over the next month or two, here's what success looks like" Pilot: - "Start small - one project, one dataset" - "Learn the workflow without disrupting everything" - "Get your team comfortable with the tools" Integrate: - "Connect OME-NGFF to your existing analysis" - "You don't have to rebuild everything" - "It should ENHANCE your workflow, not replace it" Measure: - "And track the benefits - time saved, errors reduced" Question: - "What would be a good pilot project in YOUR lab?" Timing: 45 seconds

SPEAKER NOTES - Long-term Vision (45 seconds) Looking ahead: - "And looking 6-12 months out, imagine..." Scale: - "Datasets that seemed impossible become routine" - "Multi-terabyte analyses running in the cloud" - "Your science limited by questions, not infrastructure" Leverage AI: - "AI assistants handling the grunt work" - "You focus on interpretation and discovery" - "And you have automation that actually works" Contribute: - "Here's what's BEAUTIFUL - as you build tools and workflows" - "Share them back with the community" - "Make OME-NGFF better for everyone" - "That's how open science works" Timing: 45 seconds

SPEAKER NOTES - Python Library (45 seconds) For Python users: - "Even if you don't want to use AI assistants yet" - "If you're a Python person, we've got you covered" - "The library integrates seamlessly with your existing stack" - "The package has over 15,000 monthly downloads from PyPI and a dozen contributors" Jupyter: - "Works great in notebooks - try it interactively" - "See the examples in our docs" Ecosystem: - "Built on NumPy, works with Dask for parallel processing" - "Xarray for labeled arrays" - "Familiar tools, with OME-NGFF superpowers" Custom pipelines: - "Use it as a building block for your own workflows" - "Not locked into our way of doing things" Cross-reference: - "And speaking of cloud - Eric Perlman's talk this afternoon" - "Will go deep on cloud workflows" - "Highly recommended!" Timing: 45 seconds

SPEAKER NOTES - TypeScript Library (45 seconds) Web developers: - "We also have a TypeScript/JavaScript version" - "Build web apps, visualization tools" Browser-based: - "Run OME-NGFF analysis right in the browser" - "No server needed for many tasks" Cloud deployment: - "Deploy to Vercel, Netlify, anywhere" - "Share interactive visualizations with collaborators" Timing: 45 seconds

SPEAKER NOTES - fideus Intro (30 seconds) Introduce fideus: - "Quick word about who we are" - "We're fideus labs - this is our passion" What we do: - "We specialize in biomedical imaging infrastructure" - "We're active contributors to OME-NGFF and related tools" - "We're committed to open science and open source" Not a sales pitch: - "We're here to help the community" - "Whether that's with us or on your own" - "Happy to chat after the talk or during breaks" Timing: 30 seconds - brief and humble

SPEAKER NOTES - Key Takeaways (1 minute) Summarize the message: - "Let me leave you with four key points" Read each one with emphasis: 1. "OME-NGFF:" - "This format was DESIGNED for the AI era" - "Cloud-ready, open, future-proof" 2. "MCP:" - "The bridge that makes AI assistants actually useful" - "Standard protocol, works everywhere" 3. "ngff-zarr:" - "Tools you can use TODAY" - "Free, open source, ready to go" 4. Your impact: - "This isn't just about convenience" - "It's about MULTIPLYING your scientific impact" - "Better reproducibility, broader collaboration, faster discovery" Pause: - "These four things together are transformative" Timing: 1 minute - should be at ~24 minutes total

SPEAKER NOTES - Questions & Discussion (1+ minute, flexible) Final thoughts: - "That's what I wanted to share with you today" - "We should have a few minutes for questions" - Check actual time - adjust accordingly Open the floor: - "What questions do you have?" - "Anyone have a specific use case you want to discuss?" - "How much time is a postdoc going to spend on the next 50 images?" Be ready for common questions: - Performance compared to other formats - Existing tool compatibility - Learning curve for teams - Cloud storage costs - Production readiness Encourage engagement: - "Don't be shy - if you're wondering it, others are too" - "I'll also be around during breaks if you want to chat more" Thank them: - "Thank you for your attention" - "Excited to see what you build with these tools" - "Let's make bioimage analysis better together" Timing: Flexible - use remaining time, but don't run over!

SPEAKER NOTES - Resources (45 seconds) Point to resources: - "As we go through questions and discussion, everything I've shown you is documented here" ngff-zarr docs: - "Start here - we have tutorials, examples, API docs" - "Including documentation on the ngff-zarr MCP server setup" OME-NGFF spec: - "For the deep dive into the format specification" - "This is the authoritative source" Community: - "Join the OME community - forums, mailing lists" - "People are friendly and helpful!" Timing: 45 seconds - practical and actionable