MLOps World | GenAI Summit

LinkedIn Open-sources Greykite, a library for time-series forecasting 😎👏😮 #mlopsworld2021 github.com/linkedin/greyk…

Evals. Governance. They keep coming up in every meeting, every committee session, every conversation right now. Teams are still making decisions without clean metrics. Agents are in production with no clear answer for who owns accountability when something goes wrong.

How you can go about hiring your VP of Engineering? A nicely laid out framework here (from @Shopify 's VP of Engineering) Great share, @fnthawar review.firstround.com/hiring-a-vp-of… #mlopsworld2021

AI/Data Landscape Credit: @LFAIDataFdn Open Source AI & Tools Demo week at #MLOpsWorld2021 mlopsworld.com/tools/

We surveyed our TMLS 2026 steering committee on their biggest constraints. The clearest signal was evals. Here's part 3 of what we learned when exploring evals when operating in prod. torontomachinelearning.com/the-biggest-co…

We surveyed our TMLS 2026 steering committee on their biggest constraints. The clearest signal was evaluations. In our first committee meetings, we tried to drill down on some underlying principles. Some of what we discussed: torontomachinelearning.com/the-biggest-co…

If you run production ML (platform/infra/applied), we surveyed the MLOps World 2026 steering committee on what's getting in the way of shipping models + agents. Clearest signal: evaluation + testing. Notes from our first session: mlopsworld.com/post/the-bigge…

Last June, 60+ women across the data and AI ecosystem gathered in Toronto for a Women in AI event hosted by TMLS to name the barriers they actually face, document them honestly, and map what needs to change. Nine months of work. Here it is. 🧵

Ten years in. The feedback keeps saying the same thing. "One of the highest signal-to-noise ratios among all AI conferences." Practitioners who ship. No sales pitches. No fluff. Early Bird pricing closes March 31st → See TMLS 2026 details and register

We surveyed our TMLS 2026 steering committee on their biggest constraints. The clearest signal was evaluations. In our first committee meetings, we tried to drill down on some principles. Some of what we discussed: torontomachinelearning.com/the-biggest-co…

TMLS committee survey signal: evaluation and testing was the top blocker. We broke down what the committee told us, and how it shaped this year’s program → torontomachinelearning.com/the-biggest-co…

We just launched the first 5 issues of AI in Production Field Notes, post-deployment writeups with real architectures + metrics. Read them here: @tmlsinsights" target="_blank" rel="nofollow noopener">substack.com/@tmlsinsights

Two recurring production traps: - RAG fails from retrieval latency/cost/permissions/eval gaps, not “bad embeddings.” - If you can solve it with a workflow, agents are expensive overengineering (debugging + cost predictability collapse fast).

If you’ve ever shipped a “simple” LLM feature and it turned into latency budgets + permission bugs + an on-call problem, this is for you.

Signal from our 2026 TMLS Steering Committee: among teams running AI in prod (or serious pilots), the bottleneck usually isn’t model capability. It’s organizational readiness: ownership, approvals, and operating reality.

TMLS turns 10 this year. 🇨🇦 A decade of practitioner-shaped conversations on what holds up under Canadian constraints, across Research, Practitioner, and Executive tracks. TMLS 2026 registration is live (including Early Bird options): eventbrite.ca/e/toronto-mach…

If you’ve ever inherited a “shipped” model and then spent months paying for it in incidents… you’re the audience. MLOps World = operator-led, post-deployment lessons. 📅 Nov 17–18, 2026 • UT Etter-Harbin l Austin, Texas Save the date.

Fortune 500 AI dies in 9-month approval cycles. By the time it ships, it’s yesterday’s tech. Toyota ships production AI in ~3–4 months. One owner’s-manual project saved ~$22M/year. Difference isn’t budget or talent. It’s team structure + execution cadence.

We just launched 5 long-form Substack Field Notes, written from real production talks. Not trends. No “hot takes” What broke. What scaled poorly. What actually worked. Read: @tmlsinsights" target="_blank" rel="nofollow noopener">substack.com/@tmlsinsights

RAG “fails in prod” for the most boring reason: retrieval breaks at scale. Drift, latency/cost blowups, permission mismatches, zero observability, and the model gets blamed. 👉 Full field note: open.substack.com/pub/tmlsinsigh…