Capstone Project — Hospital AI Prompt Engineering Portfolio

** Prompt Engineering: Mastering AI Communication from Zero to Expert
Lesson Content
0% Complete

Capstone Overview

This capstone applies ALL 5 principles in a single, continuous hospital workflow.

You will follow Maria Gonzalez through her complete care journey — from ER arrival to home discharge.

Every deliverable in this capstone builds on the previous one. The patient story is the thread.

 

Time: 60–90 minutes   |   Deliverables: 5 engineered prompts + 5 AI outputs + 1 reflection page

 

The Patient — Maria Gonzalez

Full Clinical Context for the Capstone

Name (for training only — de-identify in all prompts): Maria Gonzalez

Age: 62 years old | Language: Spanish (limited English) | Setting: City General Hospital

Presenting complaint: Fatigue, increased thirst, blurred vision, lower leg swelling for 3 weeks

Key Labs: HbA1c 10.2% | Fasting glucose 287 mg/dL | eGFR 52 | BMP: K+ 5.1, Creatinine 1.6

Blood pressure: 158/94 mmHg | BMI: 31.4

Primary diagnosis: Type 2 Diabetes Mellitus — uncontrolled | Secondary: CKD Stage 2

Medications on admission: Metformin 500mg BID, Lisinopril 10mg daily

Allergies: Penicillin (rash)

Social: Lives alone, retired, uses public transport, daughter visits weekends

Concern: Anxious about kidney disease diagnosis. Asks ‘Will I need dialysis?’

 

Your 5 Deliverables — One Per Principle

#

Principle

Your Task

Quality Gate

1

Give Direction (P1)

Write Maria’s admission SOAP note. Assign the correct clinical persona. Include: Chief Complaint, Vital Signs, Labs, Assessment (primary + differential), Treatment Plan.

Score on rubric. Must reach 20/25. Identify which criterion was hardest to achieve.

2

Specify Format (P2)

Generate Maria’s discharge medication table. Renal-adjusted doses required. Columns: Drug Name | Dose (renally adjusted) | Route | Frequency | Purpose | Side Effect to Monitor.

Verify: Is Metformin dose appropriate for eGFR 52? Is table importable to pharmacy system?

3

Provide Examples (P3)

Create Maria’s diabetes self-management education sheet. Provide 2 examples (for different chronic conditions) to guide AI toward simple, empathetic, teach-back style. Bilingual: English + key Spanish terms.

Does AI match example style? Is language at Grade 6 level? Would Maria understand it?

4

Evaluate Quality (P4)

Score your SOAP note from Task 1 using the 5-criterion rubric. Identify the lowest-scoring criterion. Revise one element of your Task 1 prompt. Re-run. Show before/after scores.

Document: What changed? Which prompt revision caused the improvement?

5

Divide Labor (P5)

Chain Tasks 1–4 outputs into a final 2-page care summary: Page 1 — Clinical Summary (for the medical record). Page 2 — Patient Discharge Letter for Maria (plain language, bilingual).

Evaluate the chain: Where did quality drop? Which step required the most revision?

 

Capstone Evaluation Rubric

Criterion

Points

What Evaluators Look For

All 5 principles clearly demonstrated

20 pts

Each prompt explicitly uses its assigned principle

Clinical accuracy of AI outputs

20 pts

Diagnoses, medications, and labs are clinically correct and renally adjusted

Format quality and usability

20 pts

Each output is in the correct format for its clinical purpose

HIPAA compliance and patient safety

20 pts

No real PHI used, hallucination risks identified, human review noted

Quality evaluation and iteration evidence

20 pts

Before/after rubric scores shown; prompt revision documented

TOTAL

100 pts

Score below 80 = revise and resubmit

 

Submission Checklist

  1. Task 1: Your exact prompt + AI output (SOAP Note) + rubric score
  2. Task 2: Your exact prompt + AI output (Medication Table) + renal dose verification note
  3. Task 3: Your exact prompt (with examples shown) + AI output (Education Sheet)
  4. Task 4: Before prompt + before score | Revised prompt + after score | Your analysis (what changed and why)
  5. Task 5: Final 2-page care summary (Clinical Summary + Patient Discharge Letter)
  6. Reflection page (max 1 page): Which principle had the highest impact on output quality? Where did the chain break down? What would you change for real clinical deployment?

 

Capstone Connection to Real Clinical Work

Every document in this capstone maps to a real artifact used in hospital Internal Medicine:

Task 1 SOAP Note → EHR admission note

Task 2 Medication Table → Pharmacy reconciliation / dispensing import

Task 3 Education Sheet → Patient teach-back material / discharge packet

Task 4 Quality Evaluation → Clinical documentation audit / QA process

Task 5 Care Summary + Letter → Medical record summary + patient take-home packet

Course Outline