Hi, I'm Mansoor.

I study computer science at UC Berkeley in the EECS Honors program and graduate in 2027. Most of what I build sits low in the stack: networking, databases, compilers, and code that has to be both fast and correct.

This past summer I interned on the Traffic Platform team at Databricks, working on network interface attachment for GCP VMs inside an internal network control plane, in Go and Rust. The summer before, I was an SDE intern at Amazon, where I built a real-time occupancy monitoring platform for IoT sensors on AWS.

I'm looking for a new grad role starting in 2027, ideally on distributed systems or infrastructure. Email is the best way to reach me: mansoormmamnoon@berkeley.edu. I'm also on GitHub and LinkedIn, and my resume is here as a PDF.

Things I've built

vxlan-tracer finds VXLAN MTU blackholes, the bug where small pings get through but large packets disappear and nothing is logged. It's a Go CLI that attaches four eBPF programs at once and tells you which of six things is going wrong.

LLMFirewall sits between an LLM agent and its MCP tools and stops prompt injection from turning into real tool calls. Across 4,200+ attack cases it cut the attack success rate from 100% to 48%, and to 0% for attacks that try to use write tools. It's on PyPI as llmfirewall.

A matching engine in C++20, the core of an exchange. It handles 20.7 million messages a second on one thread in a synthetic benchmark, with a median latency of 0.04 µs, and replays deterministically after a crash.

Edge Deployer is a desktop app for writing and deploying serverless functions to six providers (Cloudflare, AWS Lambda, Vercel, Netlify, Fly.io, Railway) without switching between their CLIs and dashboards.

There's also an offline reinforcement learning framework I wrote to test whether a policy trained on logs is safe to deploy. More detail on all of these is on the projects page.

Where I've worked

• Databricks, software engineering intern on Traffic Platform. May–Aug 2026.
• BAIR AUTOLab, undergraduate researcher on Robo-DM, a toolkit for large robot learning datasets. Sep–Dec 2025.
• Amazon, SDE intern. May–Aug 2025.
• CS 61C at Berkeley, senior mentor and content lead. I write the weekly worksheets on RISC-V, C, caches, and pipelining. 2025–now.

The experience page has the full list, including teaching and awards.

Notes

I keep short notes on things I'm learning, most recently on how Envoy's xDS gets config to the data plane, ARIES recovery, and lock-free ring buffers.