Intelligent Financial Analytics Platform

A personal finance analytics backend that ingests transaction data, cleans and categorizes transactions, and exposes API endpoints for dashboard-ready spending insights.

Highlights

• FastAPI backend for transaction ingestion, mock Plaid-style sync, and spending analytics
• SQLAlchemy data model that runs locally with SQLite and is PostgreSQL-ready through Docker Compose
• pandas and NumPy analytics for category, merchant, monthly, and anomaly summaries
• Lightweight dashboard endpoint for demo-ready insights
• Test suite covering analytics functions and API behavior

Why This Project

This project is useful for software engineering, data analyst, and data science internships because it combines API design, database modeling, data cleaning, and product analytics.

Tech Stack

• Python
• FastAPI
• SQLAlchemy
• PostgreSQL-ready data model
• pandas + NumPy
• Plaid Sandbox-inspired mock data

MVP Features

• Transaction ingestion endpoint
• Mock Plaid sync endpoint for sandbox-style transactions
• Category, merchant, and monthly spending summaries
• Simple anomaly detection for unusually large transactions
• Dashboard summary endpoint for frontend-ready metrics
• Lightweight local dashboard at /dashboard
• Isolated test suite for analytics and API behavior

Resume Bullets

• Developed a backend analytics platform that ingests transaction data, normalizes financial records, and stores structured user spending data with SQLAlchemy.
• Implemented transaction cleaning and category-mapping workflows with pandas and NumPy to analyze monthly spending patterns across merchants and categories.
• Built REST API endpoints for spending summaries, anomaly detection, and category-level insights, enabling dashboard-ready financial reporting.

Local Run

Recommended Python version: 3.12.

python3.12 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload

Open:

• API docs: http://localhost:8000/docs
• Health check: http://localhost:8000/health
• Dashboard: http://localhost:8000/dashboard

Seed demo data:

curl -X POST http://localhost:8000/seed

Run tests:

pip install -r requirements-dev.txt
python -m pytest

API Endpoints

• POST /transactions creates a transaction
• POST /plaid/sync/mock syncs Plaid Sandbox-inspired mock transactions
• POST /seed loads Plaid Sandbox-inspired mock transactions
• GET /transactions lists transactions
• GET /analytics/category-summary groups spend by category
• GET /analytics/merchant-summary groups spend by merchant
• GET /analytics/monthly-summary groups spend by month
• GET /analytics/anomalies flags unusually large transactions
• GET /analytics/dashboard returns dashboard-ready metrics

See docs/api.md for endpoint descriptions and sample request payloads.

Environment Notes

The data stack is pinned for Python 3.12 because pandas and NumPy wheels are more reliable there than on bleeding-edge Python versions. If your default python3 is newer, create the virtual environment with a Python 3.12 binary.

The default local database is SQLite. To run with PostgreSQL:

docker compose up --build
curl -X POST http://localhost:8000/plaid/sync/mock

The Docker Compose setup uses a local Postgres container and does not require paid Plaid credentials.

Architecture

See docs/architecture.md for the data flow, analytics layer, and design tradeoffs.

Interview Talking Points

• How transaction records are normalized before analytics
• Why API validation and database modeling matter for financial data
• How pandas groupby operations support dashboard metrics
• How anomaly detection can start simple and improve with more user history

Project Roadmap

• Replace mock data with Plaid Sandbox integration
• Add user authentication
• Add recurring subscription detection
• Add React dashboard