Mansoor Mamnoon

B.A. Computer Science — EECS Honors Program  ·  GPA 3.973

Selected coursework: Operating Systems (162) · Compilers (164) · Database Systems (186) · Algorithms (170) · Machine Learning (189) · Optimization (EECS 127) · Computer Vision (180) · Probability (Data 140) · Abstract Linear Algebra (Math H110)

Software Engineering Intern — Traffic Platform

• Developed dynamic network interface attachment for GCP-provisioned VMs, enabling transparent enforcement of customer network isolation guarantees across multi-tenant cloud infrastructure.
• Implemented cloud provider integrations within an internal network control plane, interfacing with service network layers to maintain workload boundary semantics at scale.
• Designed and validated a proof-of-concept by manually provisioning and configuring VMs end-to-end, confirming architectural feasibility before full implementation.
• Collaborated with senior infrastructure engineers on architecture design, presenting implementation plans for review prior to development.

Go · Rust · GCP · Network Control Plane · Virtual Machines · Multi-tenant Infrastructure

Undergraduate Researcher

• Worked with postdoctoral researchers on extensions to Robo-DM, a data management toolkit for large-scale robot learning datasets (affiliated with UC Berkeley AUTOLab / BAIR).
• Explored efficient dataset compression, retrieval, and integration with ML training pipelines.
• Applied prior experience in real-time event pipelines to robotics data loading and systems optimization.

Python · PyTorch · Data Engineering · Robot Learning

Software Development Engineer Intern — Backend & Distributed Systems

• Architected and shipped an end-to-end, real-time occupancy monitoring platform — from ingesting live data from 100+ IoT sensors to frontend visualization — piloted across 4 Amazon buildings.
• Built a scalable, fault-tolerant backend pipeline using AWS Lambda, SQS, DynamoDB, and Timestream, processing 10K+ streaming events/min with P99 latency < 900ms.
• Automated multi-region deployment workflows via AWS CDK (TypeScript), cutting infrastructure provisioning time by 85% and enabling horizontal scaling under peak load.
• Engineered a real-time React dashboard with live occupancy visualizations (<1s data freshness), enabling operations teams to monitor usage across facilities in real time.
• Integrated CloudWatch logs, alarms, and metrics dashboards, reducing debugging time by ~40% during integration tests.
• Authored a detailed design document and executive one-pager; presented and reviewed with a panel of SDE2s and SDE3s during design review cycles.

TypeScript · AWS Lambda · ECS Fargate · DynamoDB · SQS · Timestream · CDK · React · CloudWatch

Software Engineering Intern

• Engineered a scraping and ranking system using Python, Beautiful Soup, and pandas, optimized via custom trie-based prefix matching and vectorized scoring — reducing runtime by 5× across 200+ client profiles.

Python · Beautiful Soup · pandas · Data Pipelines

eBPF VXLAN MTU Blackhole Diagnostic Tool

• Built an eBPF-based CLI in Go + C that diagnoses silent VXLAN packet-drop bugs caused by MTU mismatches — a class of issue that is invisible to standard tooling and common in k3s/Flannel and Calico deployments.
• Implemented 4 BPF programs (TC sched_cls ingress/egress + kprobes on icmp_rcv and ip_do_fragment) producing 6 diagnostic verdicts; TC attach/detach is identity-verified and rollback-safe on error.
• Shipped interfaces and collect-environment subcommands; packaged arm64 + amd64 binaries; 6/6 verdict scenarios and 20/20 TC-lifecycle coexistence cases pass on Linux 5.15 (Ubuntu 22.04).

Go · C · eBPF · Linux TC · kprobes · cilium/ebpf · netlink

MCP Security Proxy — Prompt Injection Defense

• Built a seven-layer enforcement pipeline (allowlist, lookalike detection, arg injection scan, arg sanitization, secret-flow guard, taint-aware write gating, output injection scan) in a real async stdio MCP proxy at sub-ms latency with no external model calls.
• Information-flow control with source-level taint labels categorically blocks untrusted retrieved content from authorizing write-side-effect tools regardless of detection score — cutting attack success rate from 100% to 48% overall on 4,200+ eval cases, and from 73% to 0% on capability-exploiting attacks.
• 0.6% false-positive rate. 0.16 ms average latency. 839 tests across unit + integration suites; validated against real MCP servers (filesystem, git, SQLite) across 26 scenarios. pip install llmfirewall.

Python · FastAPI · MCP Protocol · NumPy · pytest

Exchange-Style High-Frequency Matching Engine

• Built a C++20 matching engine with intrusive FIFOs, slab allocators, and cache-hot ladders, achieving 20M+ msgs/sec throughput with sub-µs median latency; validated determinism via snapshot/resume replay proof with bit-for-bit fill equivalence.
• Designed an end-to-end analytics + backtesting pipeline (Python, Parquet, Docker, Streamlit) to replay real Binance BTCUSDT data, compute microstructure metrics (spread, microprice, volatility, impact curves, order-flow autocorr), and evaluate VWAP/TWAP/POV/Iceberg strategies with reproducible PnL & risk metrics.

C++20 · Python · Docker · Streamlit · GitHub Actions

Offline Reinforcement Learning Framework for Safety-Constrained Deployment

• Built a 6-algorithm offline RL framework (BC, CQL, IQL, TD3+BC, Decision Transformer, AWAC) with a 32-dimensional traffic routing simulator (SLO constraints, incidents, diurnal patterns) and a hospital treatment simulator.
• Implemented dataset diagnostics (coverage score, behavior entropy, OOD risk), three offline policy evaluation estimators (FQE, Weighted IS, Doubly Robust) with bootstrap confidence intervals, a policy shield, and a causal failure explorer.
• CQL achieved ~72 return vs. ~51 behavioral baseline with 3% SLO violations (vs. 8% behavioral); bootstrap CI narrowed the FQE estimate to [65.1, 74.5] across 3 seeds / 50 epochs. Reproducible via make reproduce-small.

Python · PyTorch · NumPy · Streamlit · Matplotlib · pytest

Multi-Cloud Serverless IDE

• Engineered a native desktop IDE (Electron, React 19, TypeScript) to deploy serverless APIs across 6 providers — Cloudflare Workers, AWS Lambda, Vercel, Netlify, Fly.io, Railway — behind a shared 7-method deployer interface, with one-click deploy, rollback, and drift detection.
• Built a live edge runtime simulator (preview with no cloud account), IaC export (Pulumi, Terraform, Wrangler, Docker+K8s), and a sandboxed vm.Context plugin system with explicit permission manifests and 5s execution timeouts.
• AES-256-GCM secrets vault (PBKDF2, 100k iterations) that excludes credentials at the TypeScript type level; cross-platform binaries via GitHub Actions matrix builds; 85 tests across 6 suites.

Electron · React 19 · TypeScript · Pulumi · Monaco Editor · Cloudflare / AWS / Vercel APIs

Senior Mentor & Content Lead — Computer Science Mentors

• Authored weekly worksheets covering RISC-V assembly, C, memory hierarchy, caching, pipelining, and parallelism for 300+ students.
• Led small-group sections and office hours; promoted to Content Lead (Dec 2025) to drive curriculum development.

RISC-V · C · Memory Hierarchy · Pipelining · OpenMP

Course Tutor

• Led weekly tutoring groups (7 students each), hosted lab sessions for 40+ students, and held office hours serving 300+ enrolled students.
• Contributed to a library of instructional videos based on past exams; achieved above-average feedback ratings.

Python · NumPy · pandas · Statistical Analysis

Selective academic honors track recognizing top-performing CS students. Includes advanced coursework, research components, and honors thesis option.

Judged and awarded by Intuit product marketing leadership for a go-to-market proposal targeting SMBs and solopreneurs.

Qualified from 74,000+ Senior Math Challenge participants. A* in Further Mathematics A-Level.