Todd Underwood
where else to find them
what they talk about
all 6 in the filters
where they keep going back
no series more than once — all 8 in the filters
on a stage with
- Daniel Papasianonce
- Josh Hartmanonce
- Nisha Talagalaonce
all 6 in the filters
who talks about the same things
nobody close enough yet
| title | conference | date |
|---|---|---|
| Generative AI: Beyond (Just) Hype | SREcon24 Europe/Middle East/Africa | 2024-10-29 |
Hype is exaggerated but not pointless: GPT‑4, Claude, Gemini and Dolly expose real strengths and clear failures for production engineering. Autoscaling works… | ||
| Artificial Intelligence: How Much Will It Cost You? | SREcon23 Europe/Middle East/Africa | 2023-10-10 |
Generative models require huge up-front capital: accelerator hardware costs dominate (H100-class cards cost tens to hundreds of thousands each) and training a… | ||
| SRE and ML: Why It Matters | SREcon22 Europe/Middle East/Africa | 2022-10-25 |
Machine learning systems are becoming part of core infrastructure: adoption rises, models run critical paths (recommenders, fraud, autoscaling), and operations… | ||
| Panel: OpML | SREcon21 | 2021-10-12 |
Operationalizing ML faces subtle failures and distribution shifts that are hard to detect — a schema drift takes three weeks to diagnose in one recommendation… | ||
| Nothing to Recommend It: An Interactive ML Outage Fable | SREcon21 | 2021-10-12 |
An online yarn shop uses ML to rank search and surface recommendations, retraining the same model daily retraining and deploying unchanged config. Customers… | ||
| SRE for ML: The First 10 Years and the Next 10 | SREcon21 | 2021-10-12 |
Built a production ML stack for ad selection since ~2009: data ingestion and preprocessing, training cluster, a model management and pipeline orchestration… | ||
| When Good Models Go Bad: The damage caused by wayward models and how to prevent it | TWIMLcon: AI Platforms 2021 | 2021-01-19 |
Hand analysis of ~100 incidents over 10 years shows many failures are infrastructure handovers not modeling math. A large ranking pipeline with thousands of… | ||
| Automating Operations with ML | OpML '20 | 2020-07-28 |
Survey finds most ML-for-ops ideas fail in practice because labeled data is scarce, systems are dynamic, and engineering costs outweigh gains; anomaly… | ||
| How ML Breaks: A Decade of Outages for One Large ML Pipeline | OpML '20 | 2020-07-28 |
Google analyzes 96 post-mortems from a 15-year continuous pipeline running thousands of concurrent models with about 100 billion parameters, finds failures… | ||
| All of Our ML Ideas Are Bad (and We Should Feel Bad) | SREcon19 Europe/Middle East/Africa | 2019-10-02 |
Most production ML proposals fail because the data and signal aren’t there: ticket prioritization and root‑cause models lack reliable labels, suffer delayed… | ||
| PostOps: A Non-Surgical Tale of Software, Fragility, and Reliability | LISA '13 | 2013-11-03 |
Argues that ops should shrink into software: adopt PostOps by automating toil, defining SLAs, and treating services (not servers) as the unit of work… | ||
| nothing matches | ||