Uploaded May 2025 | Updated November 2026, 16 minutes ago
[2025 - Day 3 - Lightning Talks] Alexy Khrabrov shares insights from the Open Agentic AI Knowledge Stack (OAKS), exploring how Knowledge Graphs power GraphRAG and bridge enterprise AI adoption with startup innovation around structured knowledge. This talk offers valuable perspectives on organizing domain knowledge for Small Specialized Models and Domain Expert Agents, particularly relevant for teams working on structured AI architectures with graph-based agentic workflows and scalable message passing frameworks.
ABOUT THE SPEAKER:
Alexy Khrabrov, AI/ML Community Architect, Neo4j -
ποΈ GET YOUR TICKET TO AI COUNCIL 2026 ποΈ
Meet the world's top AI infrastructure minds where architects of AI share what works. Three days of high-quality technical talks and meaningful interactions.
β aicouncil.com/sf-2026
β‘ FIND US:
X: https://x.com/AICouncilConf
LinkedIn: linkedin.com/company/aicouncilconf
Website: aicouncil.com
[2025 - Day 3 - Lightning Talks] Alexy Khrabrov shares insights from the Open Agentic AI Knowledge Stack (OAKS), exploring how Knowledge Graphs power GraphRAG and bridge enterprise AI adoption with startup innovation around structured knowledge. This talk offers valuable perspectives on organizing domain knowledge for Small Specialized Models and Domain Expert Agents, particularly relevant for teams working on structured AI architectures with graph-based agentic workflows and scalable message passing frameworks.
ABOUT THE SPEAKER:
Alexy Khrabrov, AI/ML Community Architect, Neo4j -
ποΈ GET YOUR TICKET TO AI COUNCIL 2026 ποΈ
Meet the world's top AI infrastructure minds where architects of AI share what works. Three days of high-quality technical talks and meaningful interactions.
β aicouncil.com/sf-2026
β‘ FIND US:
X: https://x.com/AICouncilConf
LinkedIn: linkedin.com/company/aicouncilconf
Website: aicouncil.com
![Breaking the Proof-of-Concept Cycle: Stop Prototyping and Get Into Production | GitLab
[2026 - DAY 3 - APPLIED AI] Most organizations are experimenting with AI at work. Proof of concepts are thriving in small-scale silos, but how can you scale those wins across an enterprise? This talk discusses the symptoms that trap teams in the cycle of perpetual prototyping and provides a practical framework for breaking through common technical, organizational, and cultural barriers to scale. Learn how to plan for production from day one and turn your AI experiments into solutions that stick.
SPEAKER:
Naheil McAvinue - Director, Enterprise Analytics & Data Science, Gitlab
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf Breaking the Proof-of-Concept Cycle: Stop Prototyping and Get Into Production | GitLab](https://i.ytimg.com/vi/Wkpt2MDCaxI/mqdefault.jpg)
![Build Trustworthy LLM Apps Powered by Agentic Evals | Meta
[2026 - DAY 3 - LIGHTNING TALK] Agentic AI has certainly secured mindspace of industry. As more & more people continue to play with LLMs, they begin developing an intuition aka vibes around whether LLM is performing the task at hand well or not. While usecases like math problem, classification are easier to verify; generative content creation etc are not. Adding agents to the mix makes things lot more complex. This talk will highlight the particular challenges one may face as they continue their journey towards productionizing their Agentic application: outputs are non deterministic, ground truth is hard to find, and what used to work before no longer does. Yet LGTM isnt a deployment strategy.
While typical LLM chatbot can be seen as a more sophisticated Google search, were beginning to expect more from Agents: Dont just give an answer but actually perform the task. Security, bias, privacy etc suddenly become non-negotiable. Only after we handle these complexities, we can unblock real usecases in healthcare, finance, legal etc.
This talk tackles why agent evaluation is fundamentally harder than traditional ML testing: multi-step reasoning chains, tool use side effects & more. How to build evaluation datasets that actually reflect production scenarios, not just cherry-picked examples. Well cover automated evaluation pipelines using LLM-as-judge patterns, and when you can not avoid human in the loop. The session addresses detecting regressions before users do: setting up continuous evaluation that catches model degradation. Tricky cases when agent aces public evals but fails in production, and how to build evaluations that predict real-world performance.
SPEAKER:
Rushabh Mehta - Tech Lead, Meta
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf Build Trustworthy LLM Apps Powered by Agentic Evals | Meta](https://i.ytimg.com/vi/XZwbeuwZxSk/mqdefault.jpg)
![The Agent Attack Surface: Why AI Is Breaking Software Security As We Know It | Socket
[2026 - DAY 3 - AI SECURITY & SAFETY] For decades, software security has followed a simple mental model: developers write code, security teams review it, and vulnerabilities get patched one by one. AI agents are about to break that model entirely.
Agents dont just run code β they choose tools, write new code, and select dependencies throughout the development process, often without any human in the loop. They lack the context to make safe decisions about what theyre pulling in: whether a package is malicious, maintained, or a typosquat of something legitimate. The software supply chain becomes dramatically harder to reason about when the software itself is deciding what to include.
At the same time, AI is supercharging vulnerability discovery, finding hundreds of new flaws at a pace the current model cant absorb. Putting the burden on individual developers is no longer acceptable.
Well discuss what a new security model looks like β one that operates at the ecosystem level and responds automatically to emerging threats.
SPEAKER:
Feross Aboukhadijeh = CEO, Socket
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf The Agent Attack Surface: Why AI Is Breaking Software Security As We Know It | Socket](https://i.ytimg.com/vi/XnP3td72tng/mqdefault.jpg)
![Data Lake CDC: Are we there yet? | ClickHouse
[2026 - DAY 3 - LIGHTNING TALK] The idea of incremental reads from data lakes has been cooking for years, but few are serving it up. As a user, you must wrangle change feeds, snapshots, time travel, that one corrupted manifest file. Do you need to be a Big Data Engineer to get it right? In this lightning talk, weβll explore whatβs broken, whats just hard, and why making data lake CDC accessible is a problem worth solving.
SPEAKER:
Marta Paes - Senior Product Manager, ClickHouse
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf Data Lake CDC: Are we there yet? | ClickHouse](https://i.ytimg.com/vi/XsEBpU1lrhs/mqdefault.jpg)

![The End of the Internet As We Know It
[2026 - Day 3 - KEYNOTE] In a recent New York Times op-ed, Mozilla CTO Raffi Krikorian warned that the informal detente that kept the internet secure β where writing software and finding vulnerabilities were equally hard β is over.
Raffi sits down with Pete Soderling to go deeper: what the Mythos moment means for builders right now, how teams should think about AI as co-author rather than tool, the economics of open source in a world of expensive foundation models, and whether open models can offer a credible alternative to the frontier labs.
Read Raffis NYT essay: https://www.nytimes.com/2026/04/15/opinion/mythos-open-souce-internet.html
SPEAKERS:
Raffi Krikorian - Chief Technology Officer, Mozilla
Pete Soderling - Founder & General Partner, Zero Prime
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf The End of the Internet As We Know It](https://i.ytimg.com/vi/YCRE3yRvazg/mqdefault.jpg)
![The Mythical Agent-Month | Posit
[2026 - DAY 2 - CODING AGENTS] Code generation is converging to free. Agents run in parallel. A single developer can burn 10 billion tokens a month. So why isnt the software getting better? Drawing on Fred Brookss fifty-year-old Mythical Man-Month, this talk argues agentic engineering is repeating softwares oldest mistakes at machine speed: generating technical debt at unprecedented scale while making scope creep unstoppable. Agents can demolish accidental complexity but struggle to distinguish this from essential complexity, and they produce new accidental complexity if not kept in check. The bottleneck was never typing speed. Design taste and knowing when to say no were always the hard part. Agents are making this paradoxically even harder.
SPEAKER:
Wes McKinney - Principal Architect, Posit and creator of pandas
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf The Mythical Agent-Month | Posit](https://i.ytimg.com/vi/YQrsuZgPJ8g/mqdefault.jpg)

![No Dropped Frames: designing a VLM around a latency budget | Moondream
[2026 - Day 1 - INFERENCE SYSTEMS] Moondream is a vision language model that runs in real time on video streams. This talk covers the model-side work behind it.
Ill start with architecture: upcycling from dense to MoE, and the tradeoffs when youre optimizing for latency rather than just parameter count. Then tokenization: why we built a custom SuperBPE tokenizer and what it bought us. The goal throughout was to avoid modeling decisions that would hurt us at inference time.
Ill also cover training infrastructure. We wrote custom training engines and RL systems because existing open source projects were pushing us toward design decisions that didnt fit. Ill talk about where we diverged and what we got out of it.
Finally, inference. Real-time VLM isnt just a serving problem or a modeling problem. We built a custom inference engine alongside the model, and Ill cover how the two informed each other.
SPEAKER:
Vik Korrapati - CTO, Moondream
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf No Dropped Frames: designing a VLM around a latency budget | Moondream](https://i.ytimg.com/vi/ZTYxdekag-k/mqdefault.jpg)
![Beyond Pipelines: Introducing Pub Sub for Tables
[2025 - Day 3 - Lightning Talks] Arvind Prabhakar shares insights from reimagining data integration through Pub/Sub for Tables, exploring how data producers can publish complete, ready-to-use tables with embedded metadata and semantics. This talk offers valuable perspectives on enabling quality checks, data contracts, and observability without pipeline complexity, essential for data teams seeking to move beyond traditional record-moving approaches that strip away context and meaning.
ABOUT THE SPEAKER:
Arvind Prabhakar, Co-founder & CEO, Tabsdata -
ποΈ GET YOUR TICKET TO AI COUNCIL 2026 ποΈ
Meet the worlds top AI infrastructure minds where architects of AI share what works. Three days of high-quality technical talks and meaningful interactions.
β https://aicouncil.com/sf-2026
β‘ FIND US:
X: https://x.com/AICouncilConf
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
Website: https://aicouncil.com/ Beyond Pipelines: Introducing Pub Sub for Tables](https://i.ytimg.com/vi/ZavgAIiyyS0/mqdefault.jpg)
![Trinity: Training a 400B MoE from Scratch Without Losing Your Mind | Arcee
[2026 - DAY 3 - MODEL SYSTEMS] Training sparse Mixture-of-Experts models at scale is notoriously unstable. Experts collapse, routers drift, and loss spikes appear out of nowhere. This talk covers how we built Trinity Large, a 400B parameter MoE (13B active), trained on 17 trillion tokens with zero loss spikes.
Well walk through the decisions that actually mattered: why we replaced standard aux-loss-free balancing with a momentum-based approach (SMEBU), how interleaved local/global attention made context extension surprisingly smooth, and what broke when we first tried running Muon at scale.
Ill also cover the less glamorous stuff: our Random Sequential Document Buffer to reduce batch heterogeneity, recovering from B300 GPU faults on brand-new hardware, and the six changes we shipped at once when routing started collapsing mid-run.
Practical lessons for teams training their own MoEs or scaling up sparse architectures
SPEAKER:
Lucas Atkins - CTO, Arcee AI
π Sign up for our No BS Newsletter to get the latest technical data & AI content: https://aicouncil.com/newsletter
ABOUT AI COUNCIL:
AI Council brings together the brightest minds in data to share industry knowledge, technical architectures and best practices in building cutting edge data & AI systems and tools.
FIND US:
Website: https://aicouncil.com/
LinkedIn: https://www.linkedin.com/company/aicouncilconf/
X: https://x.com/aicouncilconf Trinity: Training a 400B MoE from Scratch Without Losing Your Mind | Arcee](https://i.ytimg.com/vi/_GXUlM5DCL4/mqdefault.jpg)