Back to top
Selected AI work

AI Prompting that turns messy problems into useful results.

I use AI as a working partner—giving it context, constraints, examples, corrections, and a clear definition of success. These projects show how I move from an initial need to a practical, polished outcome.

Real projects • Condensed prompts • Medical and personal details de-identified
01 Illustration of a recorded consultation becoming a side-by-side comparison report

Transcript synthesis · Healthcare information

Doctor Consultations Turned Into a Side-by-Side Report

My mother received complex information from different doctors. I needed a reliable way to capture what each doctor said and compare the consultations without relying on memory or scattered notes.

Using these Otter.ai transcripts of my mother’s doctor consultations, create a clear side-by-side report. Organize what each doctor said, show where their recommendations agree or differ, and keep the language understandable. Do not add medical facts that are not in the transcripts.

The recordings became a structured, readable comparison that made two long medical conversations much easier to review. It gave the family one organized reference for understanding the doctors’ positions and preparing better follow-up questions.

Otter.aiSource groundingComparative analysisPlain-language synthesis
02 Illustration of a polished responsive career website on desktop and mobile

Web strategy · Iterative prompting

A Personal Career Website Built From One Big Goal

I wanted a “hire Irene” website that brought together my AI prompting skills, technical background, leadership experience, soft skills, work history, and personality—without sounding generic or looking like a résumé pasted onto a web page.

Create a modern website about me that makes me stand out to HR and hiring managers. Show my AI prompting, web and technology experience, career results, soft skills, and hobbies. Keep refining it until it is polished, mobile-friendly, SEO-ready, easy to contact me from, and deliverable as clean standalone HTML and CSS.

The broad idea became a polished personal brand site with a clear content hierarchy, responsive design, contact information, SEO and social metadata, accessible navigation, and a clean standalone file ready for standard web hosting.

Requirements refinementUX directionResponsive HTML/CSSSEO
03 Illustration of a Word document being transformed into an online Canvas quiz

Document transformation · Education technology

A Word Final Converted Into a Canvas-Ready Quiz Workflow

Teachers needed a practical way to move an existing Word final into Canvas without rebuilding every question by hand. The workflow also had to be clear enough for another teacher to repeat.

Convert this Word final into a Canvas-importable quiz. Preserve the wording, separate the question types, make each question worth one point, identify the correct answers, and create a QTI package for import. Then turn the process into a clear teacher handout with verification steps.

The result was a repeatable Word → ChatGPT → QTI → Canvas workflow plus a step-by-step teacher handout. It reduced manual setup and included checks for wording, answer keys, point values, formatting, and testing before publication.

QTI workflowProcess designInstructional writingQuality checks
04 Illustration of an inbox being filtered into an urgent alert and a calendar action

Automation design · Information triage

An Inbox Watcher Focused on What Actually Needs Attention

I wanted an automated way to scan email for messages that require action—not just messages marked “important.” It needed to recognize several kinds of real-life urgency and separate immediate alerts from a daily overview.

Scan my email for job applications, interviews and recruiters; bills, banking, insurance and deadlines; appointments, reservations and official notices; and personal messages asking for a reply or action. Alert me to urgent items and also give me a daily summary.

The request became a scheduled triage workflow with specific attention categories, urgent notifications, and a daily digest. When an appointment surfaced, the workflow could continue from “this needs attention” to the practical next step of adding it to the calendar.

Agentic workflowEmail classificationSchedulingAction extraction
05 Illustration of body metrics, a workout log, and a progress chart

Multimodal analysis · Ongoing coaching

Fitness Screenshots and Gym Notes Turned Into a Running Coaching System

I needed more than a one-time workout answer. I wanted an ongoing coach that could read screenshots, understand shorthand gym updates, remember corrections, track machines, weights, sets and reps, and keep the advice aligned with my actual goal.

Act as my fitness coach and maintain a running gym diary from the screenshots and workout updates I share. Track each exercise, weight, set and rep; correct the log when I clarify a number; identify progress; and guide me toward losing body fat while keeping an athletic, toned look—not competition-level bodybuilding.

Unstructured messages and app screenshots became an organized training record with corrected exercise details, progress context, body-composition interpretation, and personalized next-workout guidance. The system stayed useful because the goal and the data were continually refined.

Image understandingLongitudinal trackingData normalizationPersonalization
06 Illustration of a cellular router connected to a network testing dashboard

Technical troubleshooting · Evidence synthesis

Scattered Network Tests Turned Into a Support-Ready Diagnosis

My cellular internet had inconsistent speed, latency, jitter and occasional packet loss. I needed to understand whether the pattern pointed to Wi-Fi settings, router placement, cellular signal, tower congestion, or the need for an external antenna—and explain the evidence clearly to support.

Analyze these ping results, packet-loss percentages, latency ranges, 4G/5G signal readings, router settings and placement tests. Explain what the numbers do and do not prove, recommend the next test in order, and draft a concise vendor response in my voice using the actual results.

The scattered readings became a coherent troubleshooting record. ChatGPT separated normal variation from meaningful instability, interpreted a test with 10.7% packet loss and wide latency spikes, organized the next steps, and helped present a clear evidence-based case to the provider.

Diagnostic reasoningData interpretationTest sequencingTechnical communication

The prompt is only the beginning.

My strength is turning an unclear need into a structured workflow—supplying the right context, setting constraints, checking the output, and iterating until the result is accurate, practical, and ready to use.

Talk with Irene Back to my portfolio