AI & Cloud Engineering · Quant Research & Fintech

I build AI systems and the infrastructure they run on, aimed at making capital and markets smarter. Machine learning and applied AI are the center of my work — I design and ship LLM agent systems and automation that turn messy real-world data into decisions.

Python is my daily language, I build with modern AI tooling, and I deploy through containerized, cloud-based workflows (AWS Cloud Practitioner in progress). I treat security as part of the build, not an afterthought: on-device processing, OAuth, and careful handling of tokens and secrets.

I approach quantitative research and trading as an engineering problem: form a thesis, dig for evidence, stress the downside, and let the data settle it.

Open to opportunities with quant desks, hedge funds, and cloud/AI engineering teams.

AI & Agents

Languages

Cloud & DevOps

Backend & Data

A voice remote that runs entirely on-device: speech is processed locally with Vosk and Whisper (~150 ms on GPU), keeping it private and fast. Uses OAuth PKCE with secure on-device token storage, personalized fuzzy matching against your own artists, playlists, and liked songs, and a Windows system-tray UX with a one-command installer. Zero cloud, zero cost.

Python Vosk faster-whisper OAuth PKCE spotipy

Volatility and risk primitives written from scratch in C11: GARCH(1,1) and range-based volatility estimators, Kelly and vol-target position sizing, and purged/embargoed walk-forward evaluation. Sanitizer-clean CI on Windows, macOS, and Linux.

C11 GARCH walk-forward validation GitHub Actions

A Chrome extension that spins slots to pick a YouTube video to watch while you eat. Because deciding is the hardest part of the meal.

JavaScript Chrome Extensions API

Selected private work (code not public):

Density — Autonomous quantitative trading system for prediction markets that prices contracts from probability distributions rather than opinions. Extracts probability densities from external reference markets and trades the gap through a risk layer nothing else in the system can override. Probability model calibrated on 1,076 days of history with 94.9% observed coverage on a 95% interval; the deployed strategy produced a 1.3 Sharpe ratio with an 8.6% max drawdown across 720 live trades. Includes walk-forward validation, historical fill modeling, a 112-check test suite, crash recovery, watchdog orchestration, a live dashboard, and an LLM review layer.

Docket — Multi-tenant AI platform that turns documents into structured, trustworthy data. Organizations define their own extraction schemas and get back exact fields, each with a confidence score and a citation to source text; low-confidence results route to a human. Runs as an event-driven serverless pipeline across 13 AWS services, from a Cognito-secured API to Bedrock for inference, with tenant isolation built into the data model. Zero idle cost; ships with 49 tests, credential-free CI, and five rounds of security auditing.

BlotterBrain — LLM-powered, hedge-fund-style research workflow in Python. A single ticker triggers a multi-agent research team — analysts, bull/bear researchers, research manager, trader, risk analysts, portfolio manager — that produces a full investment memo with debate logs, risk review, and a final buy/hold/sell call. Logs each thesis, benchmarks against SPY, and feeds results into future runs. Backtests produced a 37.8% annualized return with a 71.4% win rate. (Forked from TradingAgents.)

• Collaborator / Apprentice — Riffyx Labs Mentorship (Jun 2025 – Jun 2026): AWS fundamentals, debugging, and Git-based workflows (branches, pull requests); worked alongside Cisco professionals on implementation tasks.
• B.S. Computer Science — California State University San Marcos (2025 – present)

• Anthropic certified
• OpenAI — Agents and Workflows and Applied AI Foundations
• AWS Cloud Practitioner (in progress)

Looking for opportunities with quant desks, hedge funds, and cloud/AI engineering teams. Reach me on LinkedIn or at haeganm@gmail.com.