M.S. Computer Science @ Northeastern University (Dec 2026)
The bottleneck for scaling AI isn't the models — it's the infrastructure underneath.
class AndrewNguyen:
def __init__(self):
self.role = "ML Infrastructure & Network Engineer"
self.background = "First-gen student — biochemistry → CS, unconventional path into tech"
self.interests = [
"High-speed networking for AI/HPC (RDMA, RoCE, InfiniBand)",
"Distributed inference & model serving at scale",
"ML infrastructure & evaluation systems"
]
self.fun_fact = "Likes to read philosophy/psychology books and plays tennis/badminton :)"
Centene Corporation · ML Engineer Intern · Sep 2025 – Present
- Built end-to-end LLM evaluation pipeline (MLflow 3.4 + Spark SQL) — projected ~$15K/month savings via Streamlit model recommender
- Deployed GPT-OSS-20B on 4× A10G GPUs with custom Pyfunc signatures — 50% latency reduction via GPU memory persistence + bfloat16
- Benchmarked 8 chunking-embedding configs for RAG across 2 domains (Recall@K, MRR) — reduced stack selection from days → minutes
Microsoft · Software Engineer Intern · Jun 2025 – Aug 2025
- Automated config drift detection across Microsoft's WAN (thousands of devices) — 25% reduction in incident response time
- Built YAML-driven remediation pipeline (KQL → CR/ICM → config push) — eliminated 90% of manual remediation
- Tier-2 diagnostics across L2/L3 protocols (TCP/IP, DHCP, ACLs)
San Diego State University · Cloud Data Intern · Dec 2023 – Jun 2024
- 12% model accuracy improvement via feature selection & tuning
- PyTorch recommendation system on MongoDB e-commerce data, deployed via AWS Lambda + Kubernetes
- Redis caching + CI/CD pipelines (Jenkins, Docker)
Vigitron · Network Engineer · Hardware & Networking
- Hands-on with switches, cabling, routers — L2/L3 diagnostics & protocol automation
- Built networking foundation from the physical layer up
| Project | What & Why |
|---|---|
| p4-learning-gsoc2026 | In-network ML on programmable switches — Planter (SIGCOMM) decision trees compiled to P4, PTF-tested CI; GSoC 2026 w/ P4 Language Consortium |
| smart-home-security-system | Dual-mode IoT security on RPi5 — PIR motion detection, MJPEG live stream, WebSocket alerts, JWT auth; fully local, no cloud |
| rpi5-openclaw-assistant | 24/7 edge AI assistant on RPi5 — two Telegram bots: Claudius (free, Qwen3:8b on Mac via Ollama, web search) + Apollius (Claude Code CLI); Telegram → Pi → Mac routing, Tailscale mesh |
| llm-conquestfour | On-device LLM Connect Four — Apple Silicon optimized (Ollama + llama.cpp + Metal GPU, Phi-3-mini/Mistral-7B), narrative director arc control — 🥈 2nd + People's Choice / 28 teams, Qualcomm × Microsoft × NEU Hackathon |
Seeking roles (graduating Dec 2026) in ML Infrastructure, HPC/Networking, and Cloud/MLOps.
