Harshil Agrawal — Portfolio & ML Demo

A minimal FastAPI portfolio website with a deployed Random Forest microservice. Built to demonstrate production ML serving, CI/CD, and containerization.

Features

• Portfolio pages — Home, Projects, Resume, Blog, Contact
• Live ML Demo — Manufacturing Defect Prediction (Random Forest) via /predict
• Production patterns — FastAPI, Docker, and Google Cloud deployment

Stack

• FastAPI + Jinja2
• scikit-learn Random Forest
• Docker
• Google Compute Engine (Always Free)

Quick Start

# Create venv
python3 -m venv .venv
source .venv/bin/activate

# Install
pip install -r requirements.txt

# Train model
python ml/train_model.py

# Run locally
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Open http://localhost:8000

API

• GET / — Home page
• GET /projects — Projects page
• GET /demo — ML demo page
• POST /predict — Defect prediction endpoint
• GET /health — Health check
• GET /api/projects — Projects JSON

Deployment

Google Cloud Always Free

1. Create an Ubuntu e2-micro VM in us-west1, us-central1, or us-east1, and allow inbound HTTP traffic.

2. SSH to the VM, then install Docker, Docker Compose, and Git:

   sudo apt-get update
   sudo apt-get install -y docker.io docker-compose-v2 git
   sudo systemctl enable --now docker

3. Clone this repository and start the app:

   git clone https://github.com/Harshil411/portfolio.git
   cd portfolio
   sudo docker compose up -d --build

4. Confirm the service is running at http://YOUR_VM_IP/health.

The included compose.yaml restarts the portfolio after a VM reboot and serves HTTPS automatically when the configured hostname resolves to the VM. Keep the VM in the listed US regions and within the e2-micro/disk/egress free-tier limits.

Render

Render remains available for simple managed deployments, but its free service tier spins down after inactivity.

Model

Random Forest trained on 5,000 synthetic manufacturing sensor samples. Features: temperature, humidity, pressure, vibration, RPM. Accuracy ~0.88.

Next Steps

• Add blog post engine (Markdown-based)
• Add actual resume PDF download
• Add privacy-conscious analytics
• Add custom domain