Building an ML Portfolio That Actually Gets Callbacks

#The 15-second test

A recruiter spends 15 seconds on your GitHub. What do they see?

• A pinned mess of 8-month-old class assignments? ❌
• Three sharp, well-README'd projects that solve real problems? ✅

#The 3-project rule

Pin exactly 3. One should be deep, one should be broad, one should be shipped.

1. Deep: A single problem you understand end-to-end — a fine-tuned model, a paper implementation, a bespoke pipeline.
2. Broad: A full-stack ML product — model + API + UI + deploy.
3. Shipped: Anything with a live URL that someone (not you) has used.

#The README that converts

# Project Name
One-line hook. What it does, for whom.

**Live demo:** https://…
**Blog post:** https://…

![Screenshot](./cover.png)

## Why it exists
(2 short paragraphs — problem + your take)

## Results
(Concrete metrics. Numbers only.)

## Tech
• PyTorch • FastAPI • Next.js • Postgres

#Kill the resume clutter

• No "MS Word, MS Excel" on an ML resume.
• No hobby projects that lack a README.
• No "looking for opportunities" — it's implied.

#Metrics matter

"Improved accuracy" is invisible. "Improved F1 from 0.71 → 0.86 on the imbalanced production split" is a hire.

Good luck out there.