Speaker grid with track filterspeakers-grid
Use in Plus UI
npx shadcn@latest add https://blocks.plusui.com/r/speakers-grid.jsonSpeakers · 2 variants
← → to step through · + on a thumbnail to compare side by side (up to 3)
/preview/speakers-grid
Open ↗Default content
{
"eyebrow": "Speakers",
"title": "The people who put the models in the machines",
"description": "Forty-eight talks from engineers who ship inference to kiosks, cars, turbines and phones — and who will show you the numbers.",
"tracks": [
{
"id": "ai",
"label": "Applied AI"
},
{
"id": "edge",
"label": "Edge Systems"
}
],
"allLabel": "All tracks",
"filterLabel": "Filter speakers by track",
"keynoteLabel": "Keynote",
"action": {
"label": "See the full agenda",
"href": "#"
},
"speakers": [
{
"name": "Amara Okonkwo",
"role": "Head of On-Device ML",
"company": "Zephlo",
"portrait": {
"src": "https://images.unsplash.com/photo-1508002366005-75a695ee2d17?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Amara Okonkwo"
},
"track": "ai",
"keynote": true,
"talk": {
"title": "Shipping a 3B model to 40,000 kiosks without one cloud call",
"abstract": "Zephlo moved its checkout assistant off the cloud and onto the till. Amara walks through the eighteen months it took: the distillation runs that failed, the int4 kernels that didn't, and the rollout dashboard that finally made product and hardware agree on what 'fast enough' means.",
"slot": "Day 1 · 09:30 · Main stage"
},
"bio": "Leads the team that runs Zephlo's assistant on 40,000 retail devices. Previously built speech models for hearing aids."
},
{
"name": "Sofia Marchetti",
"role": "Research Scientist",
"company": "Pellora",
"portrait": {
"src": "https://images.unsplash.com/photo-1573496359142-b8d87734a5a2?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Sofia Marchetti"
},
"track": "ai",
"keynote": true,
"talk": {
"title": "Evals that survive contact with production",
"abstract": "Benchmarks told Pellora their new model was 6% better. Support tickets said otherwise. Sofia shows the eval suite they rebuilt from real failure logs, how they keep it from going stale, and the three metrics that now gate every release.",
"slot": "Day 1 · 16:30 · Main stage"
},
"bio": "Builds evaluation tooling for Pellora's document models. Co-author of the open TraceBench failure dataset."
},
{
"name": "Kenji Watanabe",
"role": "VP Infrastructure",
"company": "Ostrelo",
"portrait": {
"src": "https://images.unsplash.com/photo-1542909168-82c3e7fdca5c?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Kenji Watanabe"
},
"track": "edge",
"keynote": true,
"talk": {
"title": "The 2 ms budget: what real-time AI costs on a factory floor",
"abstract": "On a packaging line, a late prediction is a wrong prediction. Kenji breaks down where every microsecond goes between camera and actuator — sensor I/O, preprocessing, the model, the PLC — and which ones Ostrelo bought back with hardware and which with better software.",
"slot": "Day 2 · 09:30 · Main stage"
},
"bio": "Runs the edge fleet behind Ostrelo's 120 packaging lines. Has opinions about PCIe lanes he will share unprompted."
},
{
"name": "Tomasz Wierzbicki",
"role": "Principal Engineer",
"company": "Talvik",
"portrait": {
"src": "https://images.unsplash.com/photo-1535713875002-d1d0cf377fde?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Tomasz Wierzbicki"
},
"track": "edge",
"talk": {
"title": "Quantize, distill, measure: our int4 playbook for ARM boxes"
},
"bio": "Ports models to the cheapest hardware that will hold them. Maintains Talvik's open-source ARM kernel library."
},
{
"name": "Priya Raman",
"role": "ML Platform Lead",
"company": "Quillon",
"portrait": {
"src": "https://images.unsplash.com/photo-1531746020798-e6953c6e8e04?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Priya Raman"
},
"track": "ai",
"talk": {
"title": "One feature store, three runtimes: keeping training and edge in sync"
},
"bio": "Owns the platform that trains in the cloud and serves on 9,000 field devices. Former SRE, still carries the pager."
},
{
"name": "Henrik Aasen",
"role": "CTO",
"company": "Brindlebay",
"portrait": {
"src": "https://images.unsplash.com/photo-1601455763557-db1bea8a9a5a?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Henrik Aasen"
},
"track": "edge",
"talk": {
"title": "Vision at the loading dock: 9 ms inference on a moving forklift"
},
"bio": "Co-founded Brindlebay to count pallets with cameras instead of clipboards. Has driven a forklift for research purposes."
},
{
"name": "Yuna Park",
"role": "Founder",
"company": "Umberly",
"portrait": {
"src": "https://images.unsplash.com/photo-1534528741775-53994a69daeb?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Yuna Park"
},
"track": "ai",
"talk": {
"title": "Small models, big margins: pricing AI features you run yourself"
},
"bio": "Runs Umberly, which sells on-premise transcription to hospitals. Writes a monthly newsletter on inference unit economics."
},
{
"name": "Yusuf Demir",
"role": "Edge Platform Engineer",
"company": "Ampwick",
"portrait": {
"src": "https://images.unsplash.com/photo-1527980965255-d3b416303d12?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Yusuf Demir"
},
"track": "edge",
"talk": {
"title": "Chargers as inference nodes: fault prediction on 6,000 stations"
},
"bio": "Turns Ampwick's charging stations into a distributed sensor network. Previously embedded firmware for wind turbines."
},
{
"name": "Ingrid Solberg",
"role": "Product Lead, Assistants",
"company": "Wisplight",
"portrait": {
"src": "https://images.unsplash.com/photo-1554151228-14d9def656e4?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Ingrid Solberg"
},
"track": "ai",
"talk": {
"title": "When the assistant should say “I don't know” — and how to measure it"
},
"bio": "Leads assistant quality at Wisplight. Spent five years designing clinical decision tools before moving to AI products."
},
{
"name": "Marcus Bell",
"role": "Robotics Lead",
"company": "Mosswell",
"portrait": {
"src": "https://images.unsplash.com/photo-1500648767791-00dcc994a43e?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Marcus Bell"
},
"track": "edge",
"talk": {
"title": "Over-the-air updates for 400 field robots, and the one that bricked"
},
"bio": "Keeps Mosswell's greenhouse robots picking. Has a complete post-mortem for every firmware rollout since 2023."
},
{
"name": "Nadia Haddad",
"role": "Security Researcher",
"company": "Quillon",
"portrait": {
"src": "https://images.unsplash.com/photo-1573497019940-1c28c88b4f3e?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Nadia Haddad"
},
"track": "edge",
"talk": {
"title": "Extracting a model from a device in an afternoon (and stopping it)"
},
"bio": "Breaks into embedded AI products for a living, then helps the vendors close the holes. Runs the summit's capture-the-flag."
},
{
"name": "Daniel Ferreira",
"role": "Staff Engineer",
"company": "Pellora",
"portrait": {
"src": "https://images.unsplash.com/photo-1557862921-37829c790f19?auto=format&fit=crop&crop=faces&w=640&h=640&q=80",
"alt": "Portrait of Daniel Ferreira"
},
"track": "ai",
"talk": {
"title": "Retrieval on a phone: a 40 MB index that answers in 80 ms"
},
"bio": "Builds Pellora's offline search. Believes every embedding should fit in L2 cache and has the benchmarks to argue it."
}
]
}