Ages 14–17: Build a Real AI Portfolio in 10 Weeks (3 Projects + Rubrics + Publish Steps)

What “a real AI portfolio” looks like (and what colleges actually notice)

If your teen is 14–17 and interested in AI, the best move isn’t collecting certificates—it’s building a small set of finished, explainable projects. A real AI portfolio is simply a clear record of:

• A problem worth solving (even a small one)
• A dataset (where it came from, what it contains, what’s messy about it)
• A model (what you tried, what worked, what didn’t)
• Evidence (metrics, tests, examples)
• Communication (a README that explains it in plain English)
• Publication (GitHub + a simple demo page or video)

This matters for college applications because admissions readers and scholarship panels can quickly tell the difference between:

• “I followed a tutorial” and
• “I built something, evaluated it, and can explain the trade-offs.”

In this post, you’ll get a 10-week plan with three machine learning projects for teens, rubrics to self-score, and publish steps—so your student ends with a portfolio that feels legitimate, not random.

The 10-week roadmap (3 projects that build on each other)

Below is a practical schedule. The goal is not perfection—it’s three complete projects with increasing difficulty, each one published.

Week	Focus	Deliverable	Parent “check-in” question
1	Portfolio setup + AI basics	GitHub repo + portfolio folder structure + first README template	“Can you show me your repo and explain your project plan in 2 minutes?”
2–3	Project 1: Classic ML (tabular)	Working notebook/script + evaluation + short write-up	“What does your model get wrong, and why?”
4–6	Project 2: Computer Vision	Trained image classifier + confusion matrix + demo	“What images does it struggle with?”
7–9	Project 3: NLP + Responsible AI	Text model + bias/ethics notes + simple app/demo	“What are the risks if someone misuses this?”
10	Publish + polish week	Portfolio page + 3 project pages + 1-minute video per project	“If I’m a stranger, can I run your project in 10 minutes?”

A simple rule that keeps teens on track: every week ends with something visible—a chart, a README update, a demo link, or a short video.

The 3 best AI portfolio projects for college applications (with rubrics)

These are designed to be impressive but realistic in 10 weeks—perfect for families wondering how to build a teen coding portfolio that stands out.

Project 1 (Weeks 2–3): “Real-World Prediction” with Tabular Data

Why it’s good: Tabular machine learning is widely used in real jobs (health, finance, sports, operations). It’s also the fastest way to show strong fundamentals.

Pick one problem (examples):

• Predict student study-time vs. test score bands (classification)
• Predict house prices (regression)
• Predict which team wins based on match stats (classification)

Suggested datasets:

• Kaggle tabular datasets (choose one with 5k–50k rows)
• Local school-approved survey data (even 100–300 rows works if explained well)

Minimum feature set to include (make it feel “real”):

• Data cleaning (missing values, outliers)
• Baseline model (e.g., logistic regression or linear regression)
• Stronger model (random forest, gradient boosting)
• Evaluation (accuracy/F1 for classification; MAE/RMSE for regression)
• Error analysis (where predictions fail)

Rubric (score each 0–3):

• Problem clarity: Is the goal clearly stated and measurable?
• Data quality: Is cleaning documented (not hidden)?
• Modeling: Baseline + improved model included?
• Evaluation: Correct metric + interpretation in plain English?
• Insight: At least 2 takeaways (e.g., which features matter, where model fails)?

A “college-ready” target is 12/15+.

Project 2 (Weeks 4–6): Computer Vision Classifier + Demo

Why it’s good: Vision projects are easy to understand quickly, which helps reviewers. A solid demo can be shown in seconds.

Pick one problem (examples):

• Classify recyclable vs. trash from images
• Identify plant disease vs. healthy leaf
• Recognize handwritten digits but with a twist (your own images, not only MNIST)

Dataset options:

• Public datasets (Kaggle, Roboflow Universe)
• Your teen’s own photos (best for originality)—even 300–800 images total can work

What to build:

• Start with a pretrained model (transfer learning): MobileNet, ResNet
• Train/validate/test split
• Confusion matrix + sample “wrong predictions” gallery
• A lightweight demo:
  • Upload an image and get a prediction (Streamlit or Gradio)
  • Or record a 1-minute screen demo

Rubric (0–3 each):

• Data strategy: Balanced classes? Clear labeling rules?
• Training quality: Uses transfer learning; avoids overfitting (shows learning curves)
• Evaluation: Confusion matrix + per-class precision/recall (at least briefly explained)
• Robustness: Tests on new images (not from training set)
• Demo: Anyone can try it (link) or see it (video)

A “standout” addition is a short section: “How I would improve this with more data/time.”

Project 3 (Weeks 7–9): NLP Project + Responsible AI Notes

Why it’s good: Text is everywhere (support tickets, reviews, applications). Adding a Responsible AI section signals maturity—something admissions teams like.

Pick one problem (examples):

• Classify movie/product reviews (positive/negative)
• Sort school club suggestions into categories
• Detect spam vs. normal messages (use synthetic or approved data)

Build requirements:

• Clean text (lowercasing, punctuation rules, tokenization)
• Baseline: simple TF-IDF + logistic regression
• Improved model: small transformer (e.g., DistilBERT) or a fine-tuned model
• Evaluation: F1-score, confusion matrix, and examples of misclassifications

Responsible AI mini-checklist (include in README):

• Privacy: Where did the text come from? Any personal data removed?
• Bias: Are certain groups or topics misclassified more often?
• Misuse: What could go wrong if used in a school or community setting?
• Mitigation: What guardrails would you add?

Rubric (0–3 each):

• Clarity: Problem + label definitions are explicit
• Baseline vs. improved: Two approaches compared fairly
• Evaluation: Correct metrics + 5 example predictions shown
• Ethics: Responsible AI section is specific (not generic)
• Communication: README makes sense to a non-technical reader

If your teen is aiming for “best AI projects for college applications,” this third project often becomes the signature piece because it combines tech + judgment.

Publish steps: turn projects into a portfolio that feels professional

A strong build can still look weak if it’s hard to run or understand. Here’s a publish process that works for high school students.

Step 1: Use a consistent project structure (Week 1)

Inside each project repo (or subfolder), include:

• README.md (the story)
• data/ (or a link + instructions if data can’t be shared)
• notebooks/ or src/
• requirements.txt (or environment.yml)
• results/ (charts, confusion matrices, screenshots)

Step 2: Write a README that “passes the parent test”

A good README answers these questions in order:

• What problem are you solving?
• Why does it matter?
• What data did you use?
• What models did you try?
• How did you evaluate results?
• What are the limitations?
• How can someone run it?

If a parent can read it and understand the goal, an admissions reader will too.

Step 3: Add proof (not just claims)

For each project, include at least:

• One metrics table (even small)
• One chart (learning curve, feature importance, or confusion matrix)
• 5–10 example predictions (especially for NLP)

Step 4: Publish a demo (or a video)

Options that are teen-friendly:

• Streamlit Community Cloud (simple web app)
• Gradio + Hugging Face Spaces (great for ML demos)
• A 60–90 second YouTube unlisted demo (screen recording)

A demo turns “code” into “product,” which is exactly what reviewers like.

Step 5: Create a simple portfolio page (Week 10)

Keep it minimal:

• A short bio (2–3 lines)
• A “Projects” section with three cards
• For each card: 1-sentence problem, tech used, links to GitHub + demo/video

If building a website feels like too much, a polished GitHub profile + pinned repos also works.

Next Steps: the 10-week checklist to start this weekend

If you want your teen to actually finish (the hardest part), focus on momentum and weekly deliverables.

• Today (30–60 minutes):

  • Pick the three project topics (one tabular, one vision, one NLP)
  • Create a GitHub account (if needed)
  • Make a single folder called ai-portfolio-10-weeks

• This week:

  • Set a schedule: 3 sessions/week, 45–75 minutes each
  • Create README templates for all 3 projects
  • Choose datasets and write down data sources

• Every Sunday “portfolio review” (10 minutes):

  • Your teen shows: one new chart, one new metric, or one updated README section
  • Ask: “What did you try? What happened? What will you try next?”

• Week 10 goal:

  • Publish links (GitHub + demo/video)
  • Write a short reflection: “What I learned building these three projects”

If your student completes these three builds with the rubrics above, they won’t just have “projects”—they’ll have ai portfolio projects for high school students that read like real work, show growth, and are easy for colleges (and internship reviewers) to understand.