Back Aidan Liu
Stanford CARE Explorers Program · Summer 2026

Gut Instincts: A GI Cancer
Health Platform for Asian Americans

9-DAY INTENSIVE GASTROINTESTINAL CANCER RESEARCH WEB PLATFORM · FINAL DELIVERABLE
9 Days of instruction
8+ Stanford speakers
6 Prompt iterations
1 Live web deliverable
About the Program

Stanford's Center for Asian Health Research and Education

Stanford CARE (Center for Asian Health Research and Education) Explorers is a 9-day intensive program that brings together high school students to explore health disparities affecting Asian American communities. The curriculum combines epidemiology, data science, clinical medicine, and AI — with a focus on how research translates into real-world community health impact.

Each day featured lectures, workshops, and talks from Stanford faculty and clinicians working at the intersection of Asian health, AI, and medicine. Our group was assigned gastrointestinal cancer as our research topic. The final deliverable: build a functional website that communicates our research and provides a health tool for the community.

View Live Site Demo
Final Deliverable

Gut Instincts — What We Built

Gut Instincts is a full-stack gastrointestinal cancer information and risk assessment platform specifically designed for Asian American populations. It features seven distinct sections, an AI assistant, and an adaptive interface that changes based on the user's age group — all built to address the gap in GI cancer awareness and health literacy in Asian communities.

Key Design Goal

Most health platforms treat "Asian American" as a single category — hiding the vastly different cancer risk profiles across Korean, Vietnamese, Chinese, Japanese, South Asian, and Filipino populations. Gut Instincts disaggregates by subgroup throughout its risk assessment, educational content, and data visualizations.

AI Chatbot (Ask the Guide)
An AI assistant trained on GI cancer literature. Responses include source citations and explicit disclaimers to consult a medical professional. Never makes assumptions from hypotheticals.
Risk Profile Assessment
A personalized questionnaire collecting age, ethnicity (with full Asian subgroup disaggregation), diet, family history, smoking habits, and H. pylori exposure — producing a risk estimate per cancer type.
Diet Tracker + Streak
Users log meals to balance GI-protective foods against high-risk dietary patterns. A daily streak tracker encourages consistent logging, with progress saved per account.
Progress Dashboard
A personal dashboard showing past assessments (click to expand each), diet streak, and health progress over time. Accessible after creating an account.
Age-Adaptive UI (3 Modes)
Under 16: gamified interface with quests, badges, XP, and a parent dashboard. 17–65: the full standard UI. 65+: larger text, simplified layout, elderly-friendly interactions. Users can switch modes freely.
Education Hub (Learn)
Plain-language explainers of each GI cancer type with inline clickable citations. Data visualizations compare Asian subgroup incidence rates against national averages — with scroll animations throughout.
User Accounts + Data Persistence
Create an account to save assessment history, streaks, and diet logs across sessions. Guest mode available, but all data resets on refresh. Users can clear their own data at any time.
Multilingual + Subgroup Focus
Educational content available in multiple languages (Mandarin, Vietnamese, Tagalog). All risk data disaggregated by Asian subgroup — Korean, Japanese, Vietnamese, Chinese, Filipino, South Asian — not treated as one group.
How We Built It

Vibe Coding — From Research to Product

The website was built using vibe coding — describing software in plain English and having AI build it iteratively. As we learned on Day 6, the key is being specific, providing context, working one task at a time, and iterating. Below are the core prompt iterations that shaped the final product.

React / Replit Vibe Coding Iterative Prompting AI Chatbot Data Visualization Responsive Design
Prompt 1 — Foundation
Create a comprehensive, responsive gastrointestinal (GI) cancer informational and risk assessment website for ordinary English-speaking people that explicitly focuses on Asian subgroups and ethnic disaggregation. Include a multi-page UI, interactive AI chatbot, personalized diet tracker, streak tracker, and progress dashboard. Responses should be based on sources with disclaimers to talk to a medical professional.
Prompt 2 — Age-Adaptive UI + Visuals
Ask for user's age first before accessing the website — under 16 gets a gamified/child-friendly UI, 17–65 stays as-is, 65+ gets an elderly-tailored UI. Add a button to switch age groups. Add more visualizations comparing Asians vs. other races. Add scroll animations and purposeful color. Include all races in the ethnicity dropdown. Move assessment history into the dashboard tab.
Prompt 3 — Teen Features + Accounts + Polish
Make the 3 age UIs more distinct in text, visuals, and interaction. Add specific game features for under 16: quests, daily points, missions, XP, badges, streaks. Add a parent dashboard tracking teenager data. Add create account / login so user data persists — without an account, streaks and trackers reset on refresh. Add a clear/reset button. Assessment results should be formatted differently per age group. Add clickable inline citations on the Learn page. Title: Gut Instincts.
Speaker Notes

9 Days of Lectures, Research & Clinicians

Every day covered a different slice of AI, medicine, and public health. Here are my notes from each session.

Day 01 Large Language Models · Prompt Engineering · Python
Large Language Models
  • Learn language patterns from millions of texts; predict word-by-word
  • Evolution: rule-based AI → traditional ML → deep learning → LLMs
  • GPT divides words into tokens to read and process text
  • Opportunity to improve Asian healthcare — Asians are 60% of global pop but only 10% of medical researchers
  • Hallucination: a model that is confidently producing something false
Prompt Engineering + Fact Checking
  • Prompt engineering: design/refine instructions so AI gives a better response
  • Key elements: task clarity, context, constraints, success criteria, iteration
  • LLMs are unreliable due to hallucinations, overconfidence, outdated knowledge, bias, and vagueness
Python Basics
  • Lists: [], for loops: for i in range(3):, while loops: while i < 5:
Day 02 EDA · Disease Risk Measures · Dr. Palaniappan
Exploratory Data Analysis
  • Analyze/investigate datasets before applying formal models; summarize using visuals, stats, patterns
  • Data stored as tabular data / dataframes; use the pandas library in Python
Measures of Disease Risk
  • Prevalence: (Total cases / Total Population) × 100 — e.g., 7.5% asthma prevalence
  • Incidence: new cases over time — e.g., 9.11 COVID cases per 100 person-years
  • Relative Risk (RR): RR < 1 = protective, RR = 1 = no association, RR > 1 = increased risk
  • Odds Ratio (OR): OR > 1 = positive association — e.g., smokers are 1.62× more likely to be cases
  • P-value: probability of observed results if null hypothesis is true; reject H₀ if p < 0.05
  • Chi-square test: tests if two categorical variables are associated by chance
Dr. Palaniappan — Asian Health Disparities
  • Asians are 7% of U.S., 30% of Bay Area, but receive only 0.17% of NIH funding
  • Europeans are overrepresented in research 10×; Asians underrepresented 6× — decisions made from non-Asian data
  • CVD: mortality rates vary sharply — 24% Korea, 32% U.S., 43% China; stroke risk rises as Asians move to the U.S.
  • Cancer: 2% liver cancer in Vietnam → 16% in Vietnamese Americans
  • Diabetes: people with normal-weight diabetes have 2× the mortality of overweight/obese; strength training is best for Asians
Day 03 Statistics · Dr. Ravi · Dr. Jison Hong (Alcohol & ALDH2)
Statistics Fundamentals
  • Variables: categorical (distinct groups) vs. continuous (infinite values on a scale)
  • Central tendency: mean (symmetric data), median (skewed data / outliers)
  • Spread: range (sensitive to outliers), standard deviation (deviation from the mean)
Dr. Ravi — Digital Health
Dr. Jison Hong — Alcohol & Asian Health
  • WHO: alcohol → 3 million deaths globally; drinking can cause gastrointestinal and hepatic diseases, neurologic disorders, immunologic issues, sleep disorders
  • ALDH2 deficiency (Asian flush): causes accumulation of toxic acetaldehyde, raising cancer risk — distinct from alcohol allergy (hives, wheezing, reaction to ethanol itself)
  • Community outreach: alcohol patch tests, educational materials via AICAP program
  • My question: Do you think AI should be integrated into patient education or clinical prescriptions? — Varies by LLM type; Open Evidence is an example; responses vary significantly by model
Day 04 Correlation · Data Visualization · Dr. Chen (AI in Medicine)
Correlation
  • Correlation coefficient r: -1 to +1; 0 = uncorrelated, +1 = perfect positive, -1 = perfect negative
  • Key rule: correlation doesn't imply causation, but causation always implies correlation
Data Visualization Principles
  • Distribution → Histogram; Change over time → Line chart; Comparison → Bar chart; Relationships → Scatterplot
  • Accurately conveys the message; labeled axes/titles; minimized clutter; meaningful colors; legends when needed
Dr. Chen — AI in Clinical Practice
  • Clinician = competence (knowledge → judgment), communication (empathy → influence), character (liability → responsibility)
  • LLM accuracy: 90% on multiple choice; diagnostic reasoning: physicians + GPT < GPT alone < o1-preview
  • Management cases: physicians < physicians + GPT — AI augments human judgment
  • RAG (retrieval-augmented generation) + multi-agent perspectives significantly reduce hallucination harm
  • AI scored higher on empathy and patient confidence vs. humans in remote settings
  • Use AI in high-stakes conversations in low-stakes simulated environments first
  • My question: With RAG/cross-checking, would diagnostic reasoning studies show different results?
Day 05 Medical School & Residency Panel
  • Pre-med is not a major — you choose your major and take pre-med requirements on top
  • Residency (2–7 years) requires applications with interviews, similar to college admissions
  • Intern year = first year of residency; resident / senior resident = 2+ years
  • Specialties vary widely: surgery vs. non-surgery, pediatrics vs. geriatrics, patient-facing vs. radiology
  • Advice: don't focus on kicking the can down the road — be present and stay in the moment
  • My questions: What's your most impactful research? Advice for high schoolers considering biology/medicine? Habits for stress?
Day 06 Vibe Coding · Dr. Bryant Lin (Lung Cancer & Anemos)
Vibe Coding
  • Describing software in plain English and having AI build it for you
  • Keys: be specific, give context, one task at a time, iterate
  • Best for: data analysis, visualizations, health datasets, Python work
  • Critiques: confidently wrong (runs fine, gives incorrect results); missing edge cases; insecure (never paste real patient data into public AI)
Dr. Bryant Lin — Lung Cancer & Patient Perspective
  • Lung cancer is the #1 cause of cancer-related deaths; often not diagnosed until advanced stage
  • Chemotherapy targets all rapidly dividing cells — can kill nerves, causing numbness
  • Anemos Atlas: a large cell bank combined with AI/in silico models addressing idealized cell models that don't reflect real-world patients with diverse disease biology
  • Dr. Lin was diagnosed with never-smoker lung cancer — the exact disease he was researching; teaches From Diagnosis to Dialogue sharing his experience
  • My question: How to balance efficiency of AI with the need for physical patient–doctor interaction?
Day 07 Dr. Natalie Lui (Lung Screening) · Dr. Lisa Quach (Free Clinics)
Dr. Natalie Lui — Lung Cancer Screening
  • LDCT (low-dose CT) is used to detect lung cancer early, before symptoms appear
  • Eligibility: 20+ pack-years of smoking history
  • Only 18% of eligible people are actually screened; over 70% don't know it's available
  • Taiwan TALENT Study: family history is the single biggest risk factor (2.7% detection rate)
  • Risk factors: secondhand smoke, occupational exposures, air pollution, cooking fumes, genetic factors
  • Adenocarcinoma is more common in non-smokers; ground-glass nodules are particularly common in Asian women
  • Currently working on the LUNA study at Stanford
Dr. Lisa Quach — Cardinal Free Clinics
  • Free clinics fill gaps for the medically underinsured — privately funded, run by volunteers, free tests and medications
  • Cardinal Free Clinics serves high schoolers and above; proactive (not reactive) model
  • PFC structure: health assessment interview (undergrad/med student) + universal screening
  • AI role: help identify trends over time systematically across patient populations
  • Advice: follow your gut; it's okay to slightly digress from your general direction
Day 08 Dr. Fanglin Zhang — Stroke, Brain Science & AI Imaging
  • Stroke: blood supply to the brain is cut off; #2 cause of death globally, #1 cause of long-term disability
  • A stroke occurs every 40 seconds in the U.S.; someone dies from stroke every ~3.5 minutes
  • Types: ischemic (clot, 85%) — thrombotic or embolic; hemorrhagic (bleed, 15%) — intracerebral hemorrhage
  • AAPI populations: 2× the bleeding stroke rate of Western populations; 18.9% of AAPI strokes are hemorrhagic bleeds
  • Health disparity: language barriers, limited access to specialists, cultural factors affect treatment timing
  • BEFAST: Balance, Eyes, Face, Arm, Speech, Time — signs of stroke
  • AI saves ~52 minutes reading CT scans; brain imaging and EEG analysis for seizure detection
  • Treatment that fits one community may completely miss another — disaggregation matters
Day 09 Dr. Robert Huang — GI Cancer, Screening & Career Advice
GI Cancer Research
  • Why GI cancer for Asian Americans: different cancer burden distribution, unique risk factors, and effective interventions exist through therapy and early detection
  • Asian flush = ALDH2 deficiency: alcohol (ADH) is broken down quickly into acetate by ALDH2, but deficiency causes toxic acetaldehyde accumulation — raises cancer risk
  • South Korea's national gastric cancer screening program has improved outcomes in Korea but screening hasn't been adopted in the U.S. — similar high-risk populations go unscreened
Career Advice
  • English and communication skills matter as much or more than statistics
  • You are your own best investment — develop a unique skill that makes you marketable
  • Importance of communication: their success is your success; the best bosses could do your job
  • Make the job you have a more perfect job — add a little risk
  • A disproportionate amount of the spoils go to the top 1% in creative fields — create value which perpetuates value
  • My question: What strategies did you use to find mentors who furthered your interest in Asian health disparities? — Find people who have achieved what you want and who are motivated to stay for the long run
Group Research

GI Cancer in Asian American Populations

Our group was assigned gastrointestinal cancer. GI cancer is a group of cancers affecting the digestive system — including stomach, liver, colon, esophagus, pancreas, and gallbladder. It's the third most-diagnosed cancer type, and one of the hardest to detect early because symptoms can seem minor.

Key Finding: The Aggregation Problem

Most U.S. health surveillance reports cancer statistics for "Asian Americans" as one aggregate category. Because East Asian, Southeast Asian, and South Asian subgroups have radically different risk profiles, this aggregation both hides the highest-risk groups and creates a false impression that Asian Americans are a low-risk population for cancer screening funding.

Subgroup Primary GI Cancer Risk Key Risk Factor Relative Risk vs. White Americans
Korean Americans Gastric (stomach) cancer H. pylori + salty/preserved diet 13.3× higher gastric incidence
Vietnamese Americans Gastric + liver cancer H. pylori + Hepatitis B prevalence 6.5× higher gastric; 5–7× liver
Japanese Americans Gastric cancer H. pylori; diet; ALDH2 gene 5.2× higher gastric incidence
Chinese Americans Gastric + liver cancer H. pylori + Hepatitis B; hot beverage intake 4.8× higher gastric; 5–7× liver
South Asians (Indian/Pakistani) Gallbladder + esophageal Gallstones; betel quid chewing 2.1× higher gastric incidence
Filipino Americans Gastric cancer Diet + smoking (M: 20.6%) 1.8× higher gastric; late-stage dx
Incidence + Mortality
45–60 GI cancer cases per 100,000 Asian Americans / year
0.20% Prevalence of GI cancer among Asian Americans
12–18 Deaths per 100,000 Asian Americans annually
~60% 5-year mortality rate among diagnosed Asian Americans
What Can Be Done

South Korea's national gastric cancer screening program (endoscopy every 2 years for adults 40+, since 1999) has measurably reduced gastric cancer mortality. Japan has documented that H. pylori eradication is a cost-effective national strategy — driving an estimated 98% of gastric cancers there. For U.S. Asian American communities, targeted H. pylori test-and-treat programs (lower cost, more scalable than universal endoscopic screening) are the strongest evidence-backed lever.

Organized screening outreach — automatic mailed FIT kits, patient navigators, reminder calls, in-language materials — has closed almost the entire subgroup gap in integrated health systems, while opportunistic referral-based screening did not. Scaling that model to Korean, Filipino, and South Asian communities is the clearest next step.

Technologies & Skills
Epidemiology Biostatistics Vibe Coding Python / pandas Prompt Engineering Data Visualization AI / LLMs Public Health Research Web Development Health Disparities
View Gut Instincts Demo Stanford CARE ↗ ← Back to portfolio