Signal › Contributors

Raj Patel

AI & Infrastructure at Signal · Seattle, WA

I was a research engineer at Google Brain from 2018 to 2021, working on what was then called "ML infrastructure" and is now called "AI platform engineering." My team built internal tools for model evaluation and deployment. We were not working on the models that made headlines. We were building the plumbing that let other teams ship without breaking production.

After Google, I joined Scale AI as a senior engineer. I worked on the data engine, the system that routes labeling tasks, manages quality, and ensures that training data meets the spec. Scale was the first place I saw the full lifecycle of an AI product: from messy real-world data to a model in production to the user who sees a result and has no idea how it got there.

I left Scale in late 2023 because I wanted to write. That sounds pretentious, and maybe it is. But I'd spent five years watching AI products fail for infrastructure reasons, not model quality, not product design, but the thousand small decisions about how data flows, how models get evaluated, how inference costs compound, how monitoring works (or doesn't). Nobody was writing about this layer with any specificity.

My first article for Signal was about the real cost structure of RAG pipelines. It did well because it had actual numbers, not "it costs a lot" but "here's the per-query cost breakdown across embedding, retrieval, reranking, and generation for three different architectures at three different scale points." That's the kind of writing I care about.

I live in Seattle. I run ultramarathons, which people think is a metaphor for my work ethic but is actually just a way to stop thinking about transformer architectures for a few hours.

Experience

• Research Engineer, Google Brain
• Senior Engineer, Scale AI
• BS/MS Computer Science, Carnegie Mellon

Articles by Raj Patel (28)