Last updated: April 2026

Hi, I’m Aurelius Nguyen

I'm an honors Master's student in Computer Science at the University of Minnesota, specializing in Machine Learning and Artificial Intelligence.

I'm also passionate about Cybersecurity, Data Engineering, Quantum Computing, and Agentic Systems.

I participate in Hackathons and CTF competitions (mostly Reverse Engineering and Cryptography) every weekend.

View My WorkGet In Touch
Aurelius Nguyen
Scroll

How I approach my work

Dealt the hand. Deliver the win.

10

Recon

Study the domain, users, papers, threat models, and edge cases.

J

Decode

Break messy systems down until the core structure is clear.

Q

Forge

Build focused prototypes that turn ideas into working proof.

K

Verify

Attack it with tests, counterexamples, metrics, and failure cases.

A

Ship

Launch with observability, learn from feedback, and iterate fast.

Royal Flush! ♥

DealtDeliver

Profile

About Me

Expertise

Machine LearningMLOps on AWSFederated & Distributed LearningAutomated ReasoningNatural Language ProcessingComputer VisionAI EngineeringAgentic SystemsCryptography
Education

Integrated BS / MS in Computer Science

University of Minnesota, Twin Cities

GPA 3.8 / 4.0 · May 2023 – Expected May 2027

Coursework

CryptographyFederated LearningDeep LearningMLOps on AWSAlgorithms & Data StructuresData ScienceCyber SecurityCloud ComputingEvent-Driven & Real-Time ArchitectureParallel ComputingQuantum Computing
Experience

The Hand So Far

Latest
  1. Oct 2025 – Present

    Research Assistant - Federated Learning Security

    UMN Distributed Machine Learning Systems Lab

    Minneapolis, MN · NSF-funded ($1.1M, 3yr)

    • ▸Joined a 3-year, $1.1M NSF-funded effort (PI: Dr. Ali Anwar) on privacy-preserving federated learning systems - extending a hook-based FL testbed with pre/post-training interception via `FL_HOOK` interfaces so attacks, defenses, and config plug-ins ship without forking the core training loop.
    • ▸Implemented privacy attacks and robustness defenses as drop-in plugins. Collapsed experiment turnaround from days to hours. Lets the team explore adversarial-ML configurations that were previously infeasible.
  2. Sep 2025 – Present

    Undergraduate Research Award - ML for RSA Factorization

    UMN Distributed Machine Learning Systems Lab

    Minneapolis, MN · $2K UROP grant

    • ▸Awarded a $2K UROP grant under Dr. Ali Anwar to study neural approaches to RSA semiprime factorization - designed and trained 4 architectures (Dual-Output LSTM, Enhanced Transformer, Hybrid CNN-RNN, Factorization GAN) across 4 dataset scales (10–20 bit semiprimes) on AWS SageMaker.
    • ▸Engineered 107–117-dimensional number-theoretic feature vectors using ECPP and GNFS signals - 50–63% exact-match accuracy and a 548× lift over random chance, evidence that neural networks can recover genuine number-theoretic structure rather than overfit noise.
    • ▸First documented Transformer application to RSA factorization with multi-head attention - 52.7% exact match and 91.6% accuracy within 4-bit tolerance on a 210K-semiprime dataset.
    • ▸Designed a parameter-efficient Factorization GAN (~700K params, 4.6× smaller than the Transformer at 3.26M) that trained in 2.5 hours for 500 epochs and matched 53.7% exact-match - evidence adversarial training is competitive on mathematical-constraint satisfaction tasks.

Selected Work

Featured Projects

From classical ML to agentic AI. From research to engineering. From idea to shipped product.

Click to flip

A♠
A♠
VerificationApril 2026 – Present

D7 Agent Safety Verification Platform

A safety gate for teams deploying AI agents: turns fuzzy policy docs into machine-checkable rules so unsafe actions get blocked, with proof, before the agent ever runs them.

  • ▸30+ REST endpoints for rulesets, claims, action schemas, and verification workflows
  • ▸Z3 + deterministic SMT-LIB2 encoder; sub-50ms eval, proofs/counterexamples for high-risk requests
  • ▸Async prover workers + PostgreSQL job queues; 68+ tests across API, encoding, solver, worker layers
Code on dark monitors with crimson backlightVisit
PythonFastAPIZ3PostgreSQLDockerPydantic
K♥
K♥
ResearchDecember 2025 – Present

Zero-Knowledge Federated Learning for Medical Imaging

Lets hospitals collaboratively train medical-imaging models without trusting the aggregating server: every server-side update arrives with a proof any participating site can verify in milliseconds.

  • ▸30KB aggregation proofs with 0.3ms verification — catches weighted-update tampering
  • ▸Head-only proving on EfficientNet-B0: 5h → 59s proof generation (97% reduction); 200KB proofs, 4.6s verify
  • ▸Reproducible Flower + Docker + Kubernetes + GCP stack; cut 112s setup overhead via artifact caching
Stethoscope resting on a laptopVisit
PyTorchFlowerRustEZKLNova FoldingGCP
Q♦
Q♦
Systems2025 - 2026

L33t KV Store

A from-scratch study answering how fast a key-value store can actually go on commodity hardware once protocol and framework overhead are stripped away, browseable end-to-end so the answer is reproducible, not just claimed.

  • ▸Engineered a custom binary-protocol KV store in 400 lines of C that sustains 36,000 ops/sec at 80 microsec latency on a 3-node LAN, beating Redis 6.0 throughput by 2%.
  • ▸Drove server throughput locally from 25,000 to 185,000 ops/sec across 4 implementations spanning Python asyncio, uvloop, C with edge-triggered epoll, and C with io_uring.
  • ▸Deployed the project on Vercel as a Next.js app that runs the C protocol parser in-browser via WebAssembly.
L33t KV Store project screenshotVisit
CWebAssemblyNext.js
J♣
J♣
MLOct 2024 - Nov 2024

NeuroScan

A clinical decision-support tool for radiologists doing brain-MRI triage: shows the predicted diagnosis alongside saliency maps and an LLM rationale, so a 99%-accurate model can't sneak past audit on hidden bias.

  • ▸Engineered a production deep-learning brain tumor classifier in TensorFlow that hit >99% test accuracy across many CNN architectures (Xception, ResNet50V2) on 7,023 MRI scans spanning 4 diagnostic classes.
  • ▸Shipped a live app with Next.js and Vercel against a Dockerized FastAPI inference service on HuggingFace Spaces, delivering <3 seconds end-to-end predictions to users across 3 switchable models per request.
  • ▸Built an explainability layer pairing TensorFlow GradientTape saliency maps with Gemini 2.5 multimodal LLM analysis to catch bias in 99%-accurate models that would have failed clinical audit despite high accuracy.
NeuroScan project screenshotVisit
TensorFlowGeminiNext.js
10♠
10♠
Full-StackNov 2024 – Dec 2024

Violet — Discord Clone

A community platform for groups who want Discord-style real-time voice and chat with AI content moderation on by default, not bolted on after the first incident.

  • ▸Next.js SSR + Shadcn UI for a responsive, accessible client; WebSocket transport tuned for high concurrent loads with seamless message + voice/video sync across sessions.
  • ▸Clerk-backed multi-factor auth and Convex for real-time data + chat synchronization, eliminating manual cache invalidation and reducing race conditions in shared rooms.
  • ▸Llama-guard via Groq runs inline chat moderation, flagging policy-violating messages before they reach other users — a privacy-first content layer at the edge.
Discord logo on a blurple gradientVisit
Next.jsWebSocketClerkConvexShadcnGroq
9♥
9♥
RAGDec 2024 – Jan 2025

Nanao-chan AI Chatbot

For anyone who needs an answer from a webpage without reading it: drop in any URL, ask in plain English, get a cached RAG-backed reply in under 3 seconds.

  • ▸Puppeteer-driven dynamic scraping + Cheerio HTML parsing for any external site; Groq API powers content analysis with Axios handling clean inter-service communication.
  • ▸Upstash Redis caches scrape results and a RAG pipeline summarizes long pages — ~50% engagement lift and ~3s end-to-end response time on the average page.
  • ▸Next.js + Vercel SSR deployment for fast cold starts, scalable rollouts, and zero-config previews on every PR.
Stylized AI chatbot interfaceVisit
Next.jsRedisUpstashGroq APIPuppeteerCheerio
8♦
8♦
MLSep 2024 - Oct 2024

The Churn Report

Gives bank relationship managers a churn forecast they can act on: per-customer risk scores translated into plain-English explanations and ready-to-send retention emails, no ML knowledge required.

  • ▸Engineered a Python churn-prediction engine ensembling 8 XGBoost/sklearn classifiers across classical and SMOTE-rebalanced stacks on a 10K-customer retail-banking dataset.
  • ▸Shipped a Next.js app on Vercel against a Dockerized FastAPI service on Hugging Face Spaces, delivering sub-second predictions and a live what-if explorer that re-runs the ensemble and SHAP on every slider change.
  • ▸Paired SHAP TreeExplainer with Groq Llama 4 Scout to translate probabilities into 3-sentence risk explanations and retention emails for relationship managers with no ML expertise.
The Churn Report project screenshotVisit
XGBoostScikit-learnLlamaNext.js
7♣
7♣
AIMay 2024 – Jun 2024

Flashcard Generator

Lets students skip the busywork of building flashcards: paste any text (a chapter, lecture notes, an article) and get a study deck back in seconds, synced across devices so today's prep is on tomorrow's phone.

  • ▸Groq Llama-3.2 generates question/answer pairs from arbitrary text in <3s — handles textbook excerpts, lecture notes, and free-form prompts without prompt brittleness.
  • ▸Supabase auth + persistence for per-user decks, study history, and cross-device sync — no custom backend needed, RLS enforces per-user data isolation.
  • ▸Next.js + Vercel deployment with continuous delivery and Clerk-managed sign-in flows; minimal downtime through preview deployments per PR.
Stack of paper flashcards with Q and AVisit
Next.jsGroqSupabaseClerkLangchainOpenAI

Toolbox

Skills & Technologies

Languages and technologies I've shipped with.

PyTorch
TensorFlow
Scikit-learn
HuggingFace
Langchain
OpenAI
Pinecone
LiteLLM
Flower.ai
CUDA
Jupyter
Qiskit
AWS
Docker
Kubernetes
Terraform
Nginx
PostgreSQL
Apache Kafka
PySpark
Grafana
Prometheus
Locust
Git
Linux
Next.js
Kali Linux
IDA Pro
Ghidra
Burp Suite
Wireshark
Nmap
Python
TypeScript
JavaScript
Rust
Go
C++
Java
OCaml
SQL
R
Sage
Dafny
LaTeX
PyTorch
TensorFlow
Scikit-learn
HuggingFace
Langchain
OpenAI
Pinecone
LiteLLM
Flower.ai
CUDA
Jupyter
Qiskit
AWS
Docker
Kubernetes
Terraform
Nginx
PostgreSQL
Apache Kafka
PySpark
Grafana
Prometheus
Locust
Git
Linux
Next.js
Kali Linux
IDA Pro
Ghidra
Burp Suite
Wireshark
Nmap
Python
TypeScript
JavaScript
Rust
Go
C++
Java
OCaml
SQL
R
Sage
Dafny
LaTeX
PyTorch
TensorFlow
Scikit-learn
HuggingFace
Langchain
OpenAI
Pinecone
LiteLLM
Flower.ai
CUDA
Jupyter
Qiskit
AWS
Docker
Kubernetes
Terraform
Nginx
PostgreSQL
Apache Kafka
PySpark
Grafana
Prometheus
Locust
Git
Linux
Next.js
Kali Linux
IDA Pro
Ghidra
Burp Suite
Wireshark
Nmap
Python
TypeScript
JavaScript
Rust
Go
C++
Java
OCaml
SQL
R
Sage
Dafny
LaTeX
PyTorch
TensorFlow
Scikit-learn
HuggingFace
Langchain
OpenAI
Pinecone
LiteLLM
Flower.ai
CUDA
Jupyter
Qiskit
AWS
Docker
Kubernetes
Terraform
Nginx
PostgreSQL
Apache Kafka
PySpark
Grafana
Prometheus
Locust
Git
Linux
Next.js
Kali Linux
IDA Pro
Ghidra
Burp Suite
Wireshark
Nmap
Python
TypeScript
JavaScript
Rust
Go
C++
Java
OCaml
SQL
R
Sage
Dafny
LaTeX

ML & AI

Models, agents, retrieval, and the GPU stack underneath.

PyTorchTensorFlowScikit-learnHuggingFaceLangchainOpenAIPineconeLiteLLMFlower.aiCUDAJupyterQiskit

Infra & DevOps

Pipelines, containers, observability, and shipping pipelines.

AWSDockerKubernetesTerraformNginxPostgreSQLApache KafkaPySparkGrafanaPrometheusLocustGitLinuxNext.js

Cybersecurity

Reverse engineering, offensive tooling, and CTF muscle memory.

Kali LinuxIDA ProGhidraBurp SuiteWiresharkNmap

Languages

Daily-driven, weekend-driven, and everything in between.

PythonTypeScriptJavaScriptRustGoC++JavaOCamlSQLRSageDafnyLaTeX

Track Record

Awards & Achievements

Recognition across CTFs, hackathons, research, and merit-based scholarships.

Capture the Flag~$3K total prizes

Top placements across major CTF competitions

Reverse engineering and cryptography focus, every weekend.

Top 1 IUCTF 2025Top 3 MinneHackCTF 2025Top 4 UMNACMCTF 2025Top 18 K17CTF 2025Top 19 OlympicsCTF 2025Top 27 WannaGameCTF 2025Top 29 squ1rrel CTF 2026Top 30 BitsCTF 2025Top 31 UTCTF 2026Top 35 ECTF 2025Top 40 ApoorvCTF 2026Top 44 DawgCTF 2026Top 50 RITSEC CTF 2026Top 51 K!nd4SUS CTF 2026Top 60 pingCTF 2026Top 61 TAMUctf 2026Top 62 EHAX CTF 2026Top 71 BITSCTF 2026Top 106 IrisCTF 2024
Hackathons

Top placements at competitive hackathons

End-to-end AI and agentic products, designed, built, and shipped under weekend-tight deadlines.

Top 20 / 1,500 Headstarter HackathonParticipant @ OriginHouse 2025Participant @ MinneHack 2025
Awards & Recognition

Research grants, scholarships, and global leaderboards

Funded research, merit-based scholarships, and worldwide platform rankings.

$2K Undergraduate Research Award$60K UMN Global Excellence ScholarshipTop 94 / 118,941 @ CryptoHack.org

Competitive programming

Live ratings, write-ups, and rankings from the platforms I grind every week.

GopherHackCryptoHackLeetCode

Hand Off

Get In Touch

Open for projects, collabs, or any interesting problem. The fastest way to reach me is through the platforms below.

10♠
10♠
LocationMinneapolis, MN
J♥
J♥
Emailnguy5272@umn.edu
Q♦
Q♦
GitHubView my projects
K♣
K♣
LinkedInConnect with me
A♠
A♠
Discord_Madarame_

Built with Next.js · TypeScript · Tailwind CSS

© 2026 Aurelius Nguyen. All rights reserved.