Mansoor Mamnoon
Education
University of California, Berkeley
B.A. Computer Science, EECS Honors Program. GPA 3.973.
Experience
Databricks, 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.
- Validated the design with an end-to-end proof of concept, provisioning and configuring VMs manually before full implementation.
- Presented implementation plans to senior infrastructure engineers for architecture review before development.
BAIR AUTOLab, Undergraduate Researcher
- Worked with postdoctoral researchers on extensions to Robo-DM, a data management toolkit for large-scale robot learning datasets.
- Explored dataset compression, retrieval, and integration with ML training pipelines.
Amazon, Software Development Engineer Intern
- Built an end-to-end real-time occupancy monitoring platform, from ingesting data from 100+ IoT sensors to frontend visualization, piloted across 4 Amazon buildings.
- Built a fault-tolerant backend pipeline on AWS Lambda, SQS, DynamoDB, and Timestream processing 10K+ streaming events/min with P99 latency under 900 ms.
- Automated multi-region deployment with AWS CDK (TypeScript), cutting infrastructure provisioning time by 85%.
- Built a React dashboard with under 1 s data freshness for operations teams monitoring usage across facilities.
- Integrated CloudWatch logs, alarms, and dashboards, reducing debugging time by about 40% during integration tests.
- Wrote the design document and an executive one-pager; presented them in design reviews with SDE2s and SDE3s.
BLCK UNICRN, Software Engineering Intern
- Built a scraping and ranking system in Python with Beautiful Soup and pandas; trie-based prefix matching and vectorized scoring cut runtime 5× across 200+ client profiles.
Projects
vxlan-tracer, eBPF VXLAN MTU diagnostic
- Built an eBPF CLI in Go and C that diagnoses silent VXLAN packet drops caused by MTU mismatches, a failure invisible to standard tooling and common in k3s/Flannel and Calico deployments.
- Four BPF programs (TC ingress/egress plus kprobes on
icmp_rcvandip_do_fragment) produce one of six verdicts; TC attach and detach are identity-verified and roll back safely on error. - 6/6 verdict scenarios and 20/20 TC coexistence cases pass on Linux 5.15; ships arm64 and amd64 binaries.
LLMFirewall, MCP security proxy
- Built a seven-layer enforcement pipeline in an async stdio MCP proxy with sub-millisecond overhead and no external model calls.
- Source-level taint labels stop untrusted retrieved content from authorizing write tools regardless of detection score, cutting attack success from 100% to 48% 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 including 26 scenarios against real filesystem, git, and SQLite MCP servers. Published as
llmfirewallon PyPI.
- Built a matching engine with intrusive FIFOs, slab allocators, and cache-hot price ladders, reaching 20M+ msgs/sec with sub-microsecond median latency; snapshot/resume replay produces bit-for-bit identical fills.
- Built a Python, Parquet, and Docker pipeline that replays Binance BTCUSDT data, computes microstructure metrics, and backtests VWAP, TWAP, POV, and iceberg strategies with reproducible PnL.
- Built a framework with six offline RL algorithms (BC, CQL, IQL, TD3+BC, Decision Transformer, AWAC), a 32-dimensional traffic routing simulator, and a hospital treatment simulator.
- Added dataset diagnostics, three off-policy estimators (FQE, weighted IS, doubly robust) with bootstrap CIs, a policy shield, and a causal failure explorer.
- CQL reached about 72 return vs. about 51 for the behavior policy with 3% SLO violations (vs. 8%); FQE 95% CI [65.1, 74.5] across 3 seeds.
Edge Deployer, multi-cloud serverless IDE
- Built a desktop IDE (Electron, React 19, TypeScript) that deploys to six providers behind a shared 7-method deployer interface, with one-click deploy, rollback, and drift detection.
- Added a local edge runtime simulator, IaC export (Pulumi, Terraform, Wrangler, Docker plus Kubernetes), and a sandboxed plugin system with permission manifests.
- AES-256-GCM secrets vault (PBKDF2, 100k iterations); cross-platform builds via GitHub Actions; 85 tests across 6 suites.
Teaching
CS 61C, Senior Mentor and Content Lead
- Write weekly worksheets on RISC-V, C, memory hierarchy, caching, pipelining, and parallelism for 300+ students; promoted to Content Lead in Dec 2025.
- Lead small-group sections and office hours.
Data 8, Course Tutor
- Led tutoring groups of 7, labs for 40+ students, and office hours for 300+ enrolled students; helped produce videos on past exams.
Skills
- Languages: C++20, Go, Rust, C, Python, TypeScript, Java
- Infrastructure: AWS (Lambda, ECS, DynamoDB, SQS, Timestream), Pulumi, Docker, CDK, GitHub Actions
- ML and data: PyTorch, pandas, NumPy
- Areas: distributed systems, Linux networking, eBPF, compilers, database internals, low-latency systems
Awards
- 1st place, Intuit × Berkeley Haas Case Competition (May 2025)
- British Mathematical Olympiad qualifier, from 74,000+ UKMT Senior Mathematical Challenge participants (2022); A* in A-Level Further Mathematics