Full-Stack Software Engineer — Applied ML · Security Engineering · Software Systems
I build things end-to-end — ML inference engines in Go, RLHF data pipelines, security tooling, game engines from scratch. I don't stop at the model.
Praetorian reached out. I found their ML Binaries challenge while going through the site and decided to beat it before applying. Took two days. Then ported the full inference pipeline to Go — live on EC2, solved 11 times in a row.
Lately: working through the Hack The Box Penetration Tester path, going deeper into low-level programming, and currently reading Heavy Wizardry 101 by David Martínez Oliveira (Pico).
The direction is clear — Go, systems, security. Getting as close to the metal as possible.
500 consecutive correct. 0 wrong. Solved 11 times in a row.
| Project | What It Does | Stack |
|---|---|---|
| Binary Architecture Classifier | In-process Go inference engine classifying PE binaries by CPU architecture. Beats the | |
| Praetorian ML challenge — 500 consecutive correct, 0 wrong, 11 solves in a row. Live on EC2. | Go, React, TypeScript, AWS | |
| RLHF Eval | Data quality pipeline for RLHF training data. Seven detectors flagged 12,693 problematic preference pairs (7.9%) in | |
| Anthropic's HH-RLHF dataset. Trained competing reward models on clean vs. unfiltered data. | Python, PyTorch, FastAPI, PostgreSQL, Docker | |
| ScratchLM | GPT-2 (124M) built from scratch — custom LayerNorm, GELU, causal multi-head self-attention, transformer blocks with | |
| pre-norm residual connections. No high-level abstractions. | Python, PyTorch | |
| Crypto Microstructure Research | Statistical arbitrage research on crypto markets. ADF cointegration tests | |
| identified tradeable pairs (LINK-ADA, OP-PEPE) with 20–60 hour mean-reversion half-lives. Z-score signal generation and backtesting. | Python, pandas, statsmodels | |
| Crystal Hollows | Full 2D action-RPG demo built from scratch with zero imported assets. All rendering is programmatic via | |
Godot's _draw() API — procedural maps, multi-phase boss AI, custom dialogue system. |
Godot 4, GDScript | |
| Java HFT Engine | High-frequency trading platform — order books, matching engines, low-latency systems design. | Java |
| Project | Description | Link |
|---|---|---|
| NLWeb (Microsoft) | Identified a CI/CD gap and implemented the pipeline — Ruff linting, mypy checks, pytest matrix, Docker validation, Dependabot automation. | [PR |
| #397](nlweb-ai/NLWeb#397) |
| Languages | Go, Python, TypeScript, Java, GDScript, SQL |
| ML | PyTorch, scikit-learn, sentence-transformers, Hugging Face |
| Backend | FastAPI, REST, PostgreSQL, SQLite |
| Frontend | React, Vite, Tailwind CSS |
| Infra | AWS EC2, Docker, GitHub Actions, Linux |
-
Master of Science in Engineering — Artificial Intelligence | In Progress
-
Master of Applied Science — Software Systems | In Progress
- Bachelor of Science — Mathematics, Operations Research | Summa Cum Laude, 2024
-
ISC2 Certified in Cybersecurity | Certification Number: 2419262
-
CompTIA Security+ | In Progress
I don't do tutorials. I ship.
I use AI as a force multiplier — scaffolding, boilerplate, rubber ducking — but the architecture, the decisions, and the hard parts are mine. The goal is to become the best human programmer I can be.
Open to: Full-Stack Engineering · ML Systems · Security Engineering · Applied Research Email: anthonylewallen.dev@gmail.com