From a one-way ticket to production AI.

I was born in Jamaica.
My family left the island for the States when I was a kid.

First generation.
Nobody ahead of me had done this here —
no map, no connections, no fallback.

So I built the map myself.
Everything after this is what that cost — and what it bought.

First in my family to do this here — so I paid for it myself.
No loans, no safety net — just a car, an app, and a schedule that never quit.

I drove for DoorDash between lectures and after them —
more than 7,000 runs, semester after semester,
on a full course load at FIU.

Every tuition payment was a stack of tips.
I learned to optimize routes the way I'd later optimize models.

In my senior year, one of my personal projects caught the attention of a graduate-level professor. He saw the work — and decided to become a voice for it.

With his backing I won a $6,000 research grant and joined FIU's SOLID Lab as an Undergraduate Research Fellow (NSF REU, 2020–2021). The hard part wasn't the modeling — it was alignment. Epidemiological, economic, and market series publish at different frequencies, so I resampled every source onto a shared time index and merged on date keys before anything was comparable.

We tested pandemic severity against market indicators with correlation and regression, reporting confidence intervals alongside every point estimate — the discipline I still hold models to. It became a lead-author, peer-reviewed paper in Patterns (Cell Press), published openly with reproducible code on GitHub.

M. Hyman, C. Mark, A. Imteaj, H. Ghiaie, S. Rezapour, A. M. Sadri, M. H. Amini. Patterns, 2(8): 100315, 2021.

The professor brought me into FIU's SOLID Lab — Sustainability, Optimization, and Learning for InterDependent networks — where I built the pipelines, the model, and the public dashboard behind the paper.

After years of shipping other people's systems, I bet on my own. DeepBlueAlpha is a full-stack quantitative data platform, founded and built solo — nights and weekends, the same grind that paid for college, pointed at my own thing.

Ingestion is a batch workflow over the SEC EDGAR API, partitioning 5,000+ equities by issuer key across Azure Functions and Cosmos DB — so throughput scales horizontally and a failed run reprocesses idempotently without duplicating derived records.

Filing formats drift. XBRL tags and statement structures shift between filers and fiscal periods, so the readers have to stay backward compatible with historical filings instead of breaking on the first unexpected field.

Piotroski F, Altman Z, and Beneish M all carry traceable lineage back to the source filing, so any output can be recomputed and audited. The analytical store is separated from the operational one to serve OLAP access patterns, and end-to-end integrity checks favor correctness over timeliness. Derivative pricing (Black-Scholes, Heston, Bates) is evaluated against historical benchmarks.

It started as another after-hours personal project: one dev, one dark room. Then I stumbled onto something — instead of hand-writing a worker's rules, you could clone the worker.

Capture the attributes of an actual worker — their judgment, patterns, and decision-making — and reproduce them as an AI/ML system built in that person's image. Clone one person into many personas, each given a role — pilot, analyst, engineer, CEO — and deploy that expertise at scale. I've been building on it ever since.

Patterns (Cell Press), 2(8): 100315, 2021 — “Data analytics to evaluate the impact of infectious disease on economy.”