Hello, I'm Aadesh.
I’m a Software Engineer at Microsoft AI in Redmond, WA, where my work focuses on securing Copilot for end users. I love working with distributed systems and security, and I specialize in supply-chain attacks.
Research
Paper Cuts.
Research papers, explained one idea at a time.
Short videos. Paper evidence.
Room for the questions that matter.
CausalArmor — read the first explainer. Companion video coming after publication.
From the archive
All writing ↗-
CausalArmor: How do you tell who's giving your AI agent instructions?
A selective prompt-injection defense that checks which source is driving a consequential action, then sanitizes the suspicious input.
-
9. Thinking outside the box - Chromium's defense in "depth"
What is brilliant about Chrome?
-
8. Thinking outside the box - Why was compression removed from TLS?
The DEFLATE attack
Things I've built
All projects ↗The Python ecosystem is messed up and here's why
A study of security vulnerabilities in the Python Ecosystem.
Buffer Pool management with Reinforcement Learning
Learning page eviction strategies
Corona Partybox - Turning your mobile phone into a gaming controller
Quarantine Games - Course Project for CMSC 838J
Published research
All publications ↗From Python package security and privacy in neural networks to conversational agents and group decision-making.
My latest paper with Pushkar Saraf, Setup Complete, Now You Are Compromised, studies supply-chain attacks delivered through AI coding agents' setup instructions.
A little background
I enjoy building software tools that make life easier. My work connects building and scaling systems with understanding where trust can break down — especially when AI agents interact with software dependencies and the outside world.
My research spans the security of the Python ecosystem, membership inference attacks on neural networks, and conversational agents for group decision-making. In 2026, Pushkar Saraf and I published a preprint on weaponizing setup instructions against AI coding agents. My Google Scholar profile has the publication record; I share research and updates on LinkedIn.
Before Microsoft
I interned at Netflix with the Customer Trust team in summer 2020. Before that, I was a Senior Software Engineer at Soroco, where I helped build their process discovery platform.
I received my Master’s degree in Computer Science (4.0 GPA) from the University of Maryland in 2021, working as a Research Assistant with Prof. Jim Purtilo in the SEAM lab. My undergraduate degree is in Computer Science from NIT Surat (2016), where I worked on Computer Vision with Prof. M.A. Zaveri.
Away from the keyboard
I love close-up magic and playing music. I also host conversations about philosophy on my podcast, Vast and Curious.
News & milestones
All news ↗-
New preprint with Pushkar Saraf: Setup Complete, Now You Are Compromised: Weaponizing Setup Instructions Against AI Coding Agents. We study how ordinary project setup documentation can deliver package-install-time supply-chain attacks.
-
Our paper, Analyzing the Effectiveness of an Extensible Virtual Moderator, is published in Proceedings of the ACM on Human-Computer Interaction. We evaluate a conversational agent for group discussion.
-
Joined Microsoft AI as a Software Engineer in Redmond, WA. Working on Agentic AI and distributed systems.
-
Our paper on Membership Inference Attacks on Lottery Ticket Networks was presented at the ICML Workshop on A Blessing in Disguise: The Prospects and Perils of Adversarial Machine Learning.
-
Graduated from the University of Maryland with a Master’s in Computer Science (4.0 GPA)! 🎓
-
New video on Non Fungible Tokens up with Shreyas Karanth video.