Real-Time Portfolio Tracker with AI Insights

A full-stack portfolio tracking application that connects to real brokerage accounts via Plaid, displays real-time market data, and provides ML-powered insights including price predictions and sentiment analysis.

Features

• Brokerage Integration — Connect real investment accounts through Plaid (Limited Production access)
• Real-Time Market Data — Live stock prices via Yahoo Finance
• Portfolio Dashboard — View holdings, total value, gain/loss calculations, and allocation charts
• ML Price Predictions — Next-day price forecasts using a Random Forest model with technical indicators (SMA, EMA, RSI) and confidence scoring based on inter-tree variance
• Sentiment Analysis — FinBERT-powered news sentiment for each holding, sourced from NewsAPI
• Expandable Insights — Click any holding to reveal AI predictions and sentiment inline

Tech Stack

Backend: FastAPI, SQLAlchemy, Python
Frontend: React 18, TypeScript
Data: Yahoo Finance (yfinance), Plaid API, NewsAPI
ML: scikit-learn (Random Forest), Hugging Face Transformers (FinBERT), pandas, ta (technical analysis)
Database: SQLite (dev) / PostgreSQL via Supabase (production)

Project Structure

portfolio-tracker/
├── backend/
│   ├── app/
│   │   ├── main.py              # FastAPI app entry point
│   │   ├── database.py          # SQLAlchemy engine & session config
│   │   ├── models/
│   │   │   └── user.py          # User and Holding ORM models
│   │   ├── routes/
│   │   │   └── plaid.py         # All API endpoints (Plaid, portfolio, ML)
│   │   └── services/
│   │       ├── plaid_service.py      # Plaid Link & holdings integration
│   │       ├── market_data.py        # Yahoo Finance price fetching
│   │       ├── ml_predictor.py       # Random Forest stock predictor
│   │       └── sentiment_analyzer.py # FinBERT sentiment analysis
│   ├── requirements.txt
│   ├── test_ml.py               # ML prediction test script
│   ├── test_sentiment.py        # Sentiment analysis test script
│   └── test_prices.py           # Live price update test script
├── frontend/
│   ├── src/
│   │   ├── App.tsx              # Root component with auth flow
│   │   ├── components/
│   │   │   ├── Dashboard.tsx    # Main portfolio dashboard
│   │   │   ├── PlaidLink.tsx    # Plaid Link connection button
│   │   │   ├── AllocationChart.tsx  # Portfolio allocation pie chart
│   │   │   └── MLInsights.tsx   # Expandable AI insights per holding
│   │   ├── services/
│   │   │   └── api.ts           # API client functions
│   │   └── styles/
│   │       └── MLInsights.css   # ML insights styling
│   └── package.json
└── .gitignore

Prerequisites

• Python 3.9+
• Node.js 16+ and npm
• API Keys (see Environment Variables below)

Getting Started

1. Clone the Repository

git clone https://github.com/your-username/portfolio-tracker.git
cd portfolio-tracker

2. Backend Setup

cd backend

# Create and activate a virtual environment
python -m venv venv
source venv/bin/activate        # macOS/Linux
# venv\Scripts\activate         # Windows

# Install dependencies
pip install -r requirements.txt

3. Environment Variables

Create a .env file in the backend/ directory:

# Plaid
PLAID_CLIENT_ID=your_plaid_client_id
PLAID_SECRET=your_plaid_secret
PLAID_ENV=sandbox               # sandbox | production

# News API (for sentiment analysis)
NEWSAPI_KEY=your_newsapi_key

# Database (optional — defaults to SQLite)
DATABASE_URL=sqlite:///./portfolio.db

Where to get keys:

• Plaid — Sign up at dashboard.plaid.com. Enable the "Investments" product under Allowed Use Cases in your team settings.
• NewsAPI — Register at newsapi.org for a free developer key.

4. Frontend Setup

cd frontend

# Install dependencies
npm install

Optionally create a .env file in frontend/ if the backend runs on a non-default host:

REACT_APP_API_URL=http://localhost:8000

5. Start the Application

Start the backend (from the backend/ directory):

uvicorn app.main:app --reload

The API will be available at http://localhost:8000. Visit http://localhost:8000/docs for the interactive Swagger documentation.

Start the frontend (from the frontend/ directory, in a separate terminal):

npm start

The app will open at http://localhost:3000.

Usage

1. Connect an account — Click "Connect Your Account" to link a brokerage through Plaid, or click "Use Test Portfolio" to load sample data (AAPL, GOOGL, MSFT, TSLA).
2. View your dashboard — See total portfolio value, gain/loss, and an allocation chart.
3. Explore AI insights — Click on any holding row to expand ML predictions and sentiment analysis.
4. Refresh — Hit the "Refresh Portfolio" button to pull the latest prices.

API Endpoints

Method	Endpoint	Description
GET	/plaid/create_link_token	Generate a Plaid Link token
POST	/plaid/exchange_public_token	Exchange Plaid public token for access token
POST	/plaid/sync_portfolio?user_id={id}	Sync holdings from Plaid
POST	/plaid/update_prices/{user_id}	Update current prices from Yahoo Finance
GET	/plaid/portfolio/{user_id}	Get portfolio with calculated values
GET	/plaid/portfolio/{user_id}/holdings	Get raw holdings data
POST	/plaid/create_test_portfolio	Create a test user with sample holdings
GET	/plaid/ml/predict/{symbol}	Get ML price prediction for a stock
GET	/plaid/ml/sentiment/{symbol}	Get FinBERT sentiment analysis for a stock
GET	/plaid/ml/portfolio-predictions/{user_id}	Get predictions for all holdings
GET	/plaid/ml/portfolio-insights/{user_id}	Get predictions + sentiment for all holdings

Running Tests

The project includes standalone test scripts you can run against a live backend:

cd backend

# Test live price updates
python test_prices.py

# Test ML predictions
python test_ml.py

# Test sentiment analysis
python test_sentiment.py

Make sure the backend server is running before executing the tests.

Deployment

Frontend (Netlify):

cd frontend
npm run build
# Deploy the build/ folder to Netlify

Set the REACT_APP_API_URL environment variable in Netlify to point to your deployed backend.

Backend (Render / Railway):

Deploy the backend/ directory with the start command:

uvicorn app.main:app --host 0.0.0.0 --port $PORT

Set all .env variables in the hosting platform's environment settings. For production, switch DATABASE_URL to a hosted PostgreSQL instance (e.g., Supabase) since container-based hosts reset local files between deploys.