Jonathan G.

If the problem involves AI touching real decisions—and getting it right matters—I’m usually a good fit. I routinely deliver what would take most teams a week in a single day. As one former Upwork client put it: “Got a week’s worth of work done in less than an hour due to Jonathan’s expertise. Would work with him again, no question.” That speed comes from deep domain expertise and fluency with modern AI tooling, which I use as a force multiplier—not a shortcut. I move quickly because I know exactly what to build, what not to build, and how to avoid rework through systematic testing. Most of my work sits in environments where systems influence money, risk, or operations, and where errors are expensive. The goal is not experimentation or novelty. It is building systems that work the first time, explain their decisions, and remain reliable under scrutiny. I led the design and implementation of an AI-driven small-business credit platform for a publicly traded lender (~$40M+ annual revenue). The system automated the majority of application decisions by combining structured financial data, document understanding, and retrieval-augmented analysis with tool-enabled research. Within a controlled environment, the system could query internal data, perform constrained external searches, and assemble a 30-point underwriting profile aligned with senior credit-analyst reasoning. Crucially, the system enforced confidence thresholds and escalation logic. Only the most complex 10–15% of applications were routed to expert underwriters, allowing human effort to concentrate where judgment actually added value. Manual review fell sharply without increasing risk. Security was treated as a first-order constraint. All PII was isolated within private network boundaries, with encrypted storage and tightly scoped access. Even large-language-model interactions—including external research—were routed through secured infrastructure so sensitive data never left controlled environments. Most engagements begin with a short diagnostic phase. I’m direct about whether I’m needed at all—and when I’m not, I say so. Clients often use a less expensive developer for execution once the critical architectural and risk decisions are made. That discipline is part of the job. Selected Systems I’ve Built At Finally, a $100M Series-B payroll company, I designed and shipped a production payroll platform in roughly six weeks, making early architectural decisions that prevented later rewrites and operational debt. In academic publishing, I helped build and stabilize Refine.ink, an AI-driven peer-review platform used by faculty at leading U.S. universities, where rigor, explainability, and predictable behavior were non-negotiable. As Refine.ink’s co-founder put it: “Jonathan combines deep technical ability with unusually strong judgment. He consistently identified the right problems to solve and built systems we could trust in production.” — Ben Golub, Co-Founder, Refine.ink; former Professor of Economics and Computer Science at Harvard; current Professor of Economics at Northwestern University I’ve also built agentic and multimodal systems that execute real workflows: voice-driven legal intake systems that conduct structured interviews and route qualified matters directly into case management; and computer-vision pipelines where I fine-tuned custom object-detection models to classify industrial weld defects—often only a few pixels wide—achieving ~95% accuracy with a lightweight human-review loop. I’m typically brought in when teams need rapid progress without mistakes—when systems have to move fast and still hold up in production. I work quickly because I’m an expert at what I do and deeply fluent in AI tooling, which allows me to compress timelines without cutting corners.