Alex Rivera

The Problem

Physicians at a small health-tech startup were spending 35-45 minutes per patient on clinical notes — time taken directly from patient care. The EHR system had no AI features. The team had no ML engineers.

The Insight

The bottleneck wasn't writing — it was the blank page and the structure. Doctors know what happened. They just need to get it out fast. A model that listens to a brief voice note or rough dictation and structures it into the right clinical format removes the hardest part.

The Build

Built a Django microservice that accepts voice transcripts or rough text, sends them through GPT-4 with a clinical note prompt tuned over 6 weeks of physician feedback, and returns structured SOAP notes editable in-browser. Integrated directly into the existing patient record workflow.

The Outcome

The Problem

A regional logistics company had dispatchers working from spreadsheets and phone calls to track live shipments. No real-time visibility. Delays were discovered after they happened, not before.

The Insight

The data existed — GPS pings, driver apps, carrier APIs — but it was siloed across 4 systems with no single view. The insight was that dispatchers didn't need analytics. They needed one screen that told them what was late and why.

The Build

Built a Django + WebSocket backend that aggregated data from carrier APIs, GPS feeds, and the internal TMS. Built a React dashboard with live status, exception flagging, and one-click escalation. Went from prototype to production in 6 weeks.

The Outcome

The Problem

A credit union's month-end process involved manual data pulls, formula-heavy spreadsheets, and a reconciliation step that took two full days and always caught errors too late.

The Insight

The process wasn't complex — it was just untouched. No one had ever looked at it as a systems problem. Three scripts and a scheduler would replace two days of manual work. The barrier wasn't technical. It was that no one had asked the question.

The Build

Built a Python automation suite: data extraction from the core banking system, validation rules based on existing manual checks, and a formatted report generator that produced the month-end package automatically. Deployed on a Windows task scheduler with email alerting.

The Outcome