Isaac Martin is a software engineer at Capital One in New York, where he works on the enterprise ML feature platform — the systems that serve features to every fraud and underwriting model at the bank.
I build ML infrastructure at Capital One — feature compute engines, data-quality tooling, and SDKs used across the AI/ML division. Most recently, I built an engine that migrated 166+ features from individual to group-based compute, cutting $5.6 million per year in AWS spend.
My background bridges quantitative finance and software engineering: I studied quantitative finance at UT Austin, spent early internships in investment banking and at the FDIC, and now pursue an MS in Computer Science (Artificial Intelligence specialization) at Georgia Tech — with a 4.0 — while working full-time.
Away from the terminal I speak conversational French and am learning Japanese, play taiko drums, climb, and solve Advent of Code puzzles every December.
Enterprise ML Feature Platform (Python/SQL) serving features to all machine learning models — fraud, underwriting, and beyond.
Enterprise Architecture team (React/Go/SQL) responsible for standards and resiliency across all Capital One applications.
Internships at Focus Strategies, Navidar, and Westlake Securities (M&A analysis, comps, DCF models), the FDIC (compliance examination), and RBFCU (loan underwriting).
Built with a Hebbia engineer: a self-expanding supply-chain knowledge graph of AI-sector categories and companies mined from 10-K filings to surface second-order beneficiaries. An end-to-end pipeline ingests investment theses from Twitter and Substack, decomposes them into atomic, falsifiable claims, and verifies each against primary sources before any trade signal is generated.
Calibrated flight-cancellation odds from 3+ years of U.S. DOT on-time records. A LightGBM classifier with isotonic calibration — temporal splits, tail-number linkage to the inbound aircraft — served live with FAA SWIM streaming data, per-carrier cancellation trends, and monthly history.
A slash-command bot that scrapes a user's message history, then uses a locally-hosted LLM (Gemma 2 27B via Ollama) to generate a message in their exact writing style. A delta-syncing SQLite cache keeps re-runs fast.
/mimic @trevorSix years of December puzzle solutions across C++, JavaScript, and Python.
One page covering the ML feature platform work at Capital One, graduate coursework in artificial intelligence at Georgia Tech, AWS certifications, and selected AI projects.
Georgia Institute of Technology
MS, Computer Science — Artificial Intelligence
Expected May 2027 · GPA 4.0
CS6601 Artificial Intelligence: search, adversarial games, Bayesian networks, Gibbs sampling
The University of Texas at Austin
BBA, Quantitative Finance
May 2022 · GPA 3.91
Certificate in Applied Statistical Modeling · Texas Financial Derivatives · USIT