Hey, I'm Ayush.

I build defensive AI systems that protect people from the worst parts of the internet.

Currently in Dallas, Texas, leading AI driven threat detections at Cloudflare. Previously: Tessian (acquired by Proofpoint), Vectra AI. Once founded a healthcare AI startup called Orali Health.

LinkedIn Personal Blog Email
Ayush Kumar

A little about me

I originally wanted to go into medicine — the idea of directly helping people appealed to me. But over time I realized that technology, deployed at scale, could touch far more lives than I ever could as a single doctor. So I pursued computer engineering at UIUC and later a master's in CS at UT Austin, channeling that same impulse toward building AI systems that stop bad actors before they hurt real people.

These days I spend most of my time thinking about how to make AI security products actually useful — not just technically impressive. At Cloudflare, that means shipping things like Cloudy (an LLM that explains threats to SOC analysts in plain English) and multi-model detection systems that process north of 340 million emails a day.

Before Cloudflare, I was at Tessian building behavioral AI for email security, and at Vectra AI working on network threat detection. In a past life I co-founded Orali Health (then Vitrix Health) — we built a smartphone-based oral cancer screening device with computer vision, ran clinical trials with 3,000+ patients in Bolivia, and somehow convinced venture capitalists to fund a hardware startup run by college kids.

I believe the best security products are the ones you don't notice until they save you. And the best AI products are the ones that augment human judgment rather than replace it.

Where I've spent my time

2022 — Now

Cloudflare

Principal Product Manager, Machine Learning

I started on Cloudflare's email security team, building multi-model ML detection engines and LLM-powered security explanations. Since then I've transitioned into a broader machine learning role, creating detections that run across all of Cloudflare's products. Cloudflare sees roughly 20% of the world's internet traffic, which means the models I help ship make decisions at a scale that's genuinely hard to wrap your head around. The work spans everything from threat detection to safety classifiers to systems that catch abuse before it spreads. I still find it wild that code I helped design touches that much of the internet.

2021 — 2022

Tessian (acquired by Proofpoint, 2024)

Senior AI Product Manager

Built behavioral AI products that prevented misdirected emails, data exfiltration, and inbound threats. Tessian's whole thesis was that human error is a feature, not a bug — you can't patch people, so you build systems that understand how they actually work. I led product for internal account takeover prevention and worked through the Proofpoint acquisition strategy. Learned a lot about how enterprise security buying actually happens (spoiler: it's not rational).

2020 — 2021

Vectra AI

AI Product Manager

Network threat detection using AI. I worked on partner integrations and operational metrics, which taught me that the gap between "we detect threats" and "customers trust us" is enormous. Spent a lot of time in customer interviews with CISOs who were exhausted by false positives.

2018 — 2020

Orali Health (fka Vitrix Health)

Co-founder & CEO

Started in a UIUC entrepreneurship class with two friends. We built the ORA-1: a smartphone attachment that used high-contrast imaging and deep learning to screen for oral cancer. Won Cozad New Venture Challenge, raised from IllinoisVentures and SOSV, published in AI Med, and ran clinical trials in Bolivia. The device found cancers that standard exams missed. Ultimately didn't scale the way we hoped, but I learned more in those two years than in any job since.

Writing

I write about the messy reality of building AI security products — what works, what doesn't, and what we keep getting wrong.

Read everything on the Cloudflare blog →

Patents

A few ideas that made it through the USPTO gauntlet. Early ML work from my time at State Farm — computer vision, behavioral modeling, and the kind of applied machine learning that actually shipped.

US Patent 11,625,634 B2

System for improving user sentiment determination from social media and web site usage data

State Farm Mutual Automobile Insurance Co. · Granted Apr 2023

US Patent 10,160,457 B1

Vehicle occupant monitoring using infrared imaging

State Farm Mutual Automobile Insurance Co. · Granted Dec 2018

Now

Working on

Making AI security explanations actually useful instead of just impressive demos. Also thinking a lot about how autonomous agents can help threat intel teams without creating more chaos.

Reading

Re-reading The Obstacle Is the Way by Ryan Holiday. Reading about post-quantum cryptography and how it will reshape security. Re-reading The Almanack of Naval Ravikant because the ideas keep compounding.

Learning

Muay Thai — still getting my ass kicked, but less often than before. Doing more Kaggle competitions to keep my ML skills sharp. Rust (slowly). More about adversarial ML than I ever wanted to know.

Thinking about

Whether we're building AI security tools for the attackers we imagine or the ones that actually exist. The gap is larger than we'd like to admit.

Last updated: May 2026