PROJECTS

Software Engineer

Python & Data Projects

Projects that mirror work I’d do as a software engineer, data engineer, or data scientist: data pipelines, analytics, ML models, and service backends.

Integration Java .NET Python AI/ML Data React Angular

Navigate by software-engineering skill:

Web Apps API Services Data Pipelines ML Modeling Dashboards Testing + CLI GUI Apps
Back to Projects

More Python & data projects on GitHub

I keep extra experiments and smaller scripts in my GitHub repositories (CLI tools, notebooks, and one-off utilities).

Calorie Tracker (Django)

Full-stack Django app built in six reviewed stages, live on Vercel with a Neon Postgres backend at $0/month — auto-deploying on every push.

  • Stage 1 — Core Django: Meal model + validators, forms, FBV/CBV views, ORM daily totals, the admin.
  • Stage 2 — Users & Auth: signup/login, per-user scoping on every query, isolation tests.
  • Stage 3 — REST API: DRF ViewSet, token + session auth, Chart.js trend fed by the app's own API, CSV export.
  • Stage 4 — Production: settings split, Postgres via DATABASE_URL, Docker + healthchecks.
  • Stage 5 — API tests: committed Bruno collection — 21 assertions, headless via bru CLI.
  • Stage 6 — Serverless: Vercel Python function + Neon + CDN static — the live link above.
Python Django DRF PostgreSQL Docker Vercel Bruno pytest/Django tests
Live App

  • Stage 1 – Core Django: Meal model with validators, ModelForm, function- and class-based views, ORM aggregation for daily totals, the admin, a seed command — plus timezone-correct day boundaries (a DST edge case where Django rejects legitimate 1 AM meals was found and fixed).
  • Stage 2 – Users & auth: signup/login/logout, per-user data scoping on every query, and isolation tests that prove one user can never see another's meals.
  • Stage 3 – REST API: DRF ModelViewSet (cross-user access returns 404, not 403 — no ID enumeration), session + token auth, a throttled token endpoint, tokens revoked on password change, CSV export, and a Chart.js trend fed by the app's own /api/meals/summary/.
  • Stage 4 – Production posture: base/dev/prod settings split, Postgres via DATABASE_URL, WhiteNoise, gunicorn, non-root Docker image with healthchecks; manage.py check --deploy passes clean.
  • Stage 5 – API test collection: a committed Bruno collection (plain-text, git-friendly) — token flow, idempotent CRUD chain, negative tests; 21 assertions run headless via the bru CLI.
  • Stage 6 – Serverless: the same codebase as a Vercel Python function + Neon Postgres + CDN static files. One codebase, four run modes: runserver/SQLite, gunicorn/Postgres in Docker, and serverless — each chosen to match the workload.
  • 131 tests across stages; every stage passed an adversarial review before moving on.

Popular Baby Names Dashboard

End-to-end analysis of NYC’s Popular Baby Names. A Python prep script labels New / Rising / Falling / Stable trends, and the results are visualized in Looker Studio (or optionally a Plotly/Dash view) with interactive filters and KPI rollups.

PyCharm Pandas Looker Studio Google Sheets Plotly/Dash
Live Dashboard

Customer Metrics Pipeline & API

End-to-end churn-style pipeline: generate synthetic customer CSVs, engineer metrics, train a simple model, and expose a /score_customer FastAPI endpoint with interactive docs at /docs.

Python FastAPI Pandas scikit-learn

  • Feature pipeline over synthetic CSV customer metrics with Pandas.
  • Churn-style model trained with scikit-learn and saved for inference.
  • FastAPI service exposing /score_customer with Swagger UI and JSON responses.

Customer Churn (Cars)

Cost-aware churn model for car owners using synthetic data. TF-IDF text features plus numeric & categorical signals feed a tuned Logistic Regression classifier. Includes CLI tools (train.py, predict.py) to train and score CSV files.

Python Pandas scikit-learn TF-IDF

  • Synthetic churn dataset for car models.
  • TF-IDF vectorization and Logistic Regression baseline (GridSearchCV).
  • Cost-aware threshold selection for business-focused metrics.
  • predict.py appends label and probability to output CSVs.

Python Foundations Toolkit

Small, portfolio-style toolkit showing solid Python foundations: a reusable core package, a CLI with subcommands, a FastAPI service that reuses the same logic, and a tiny SQLite demo database seeded with synthetic rows and sample CSV files for learning and testing.

Python FastAPI SQLite CLI pytest

  • Core modules for CSV, text, and SQLite: toolkit/core/*.py.
  • CLI entry point with subcommands: python -m toolkit.cli ....
  • FastAPI app (toolkit/api.py) exposing /csv-summary and /text-summary.
  • SQLite demo DB (data/demo.db) and synthetic/sample files in data/ for practice runs.
  • pytest tests validating CSV/text helpers and core utilities.

Hello-Prod-FastAPI

Small FastAPI template organized for production: app factory pattern, versioned routes, environment-based settings, and centralized error handling.

Python FastAPI uvicorn Pydantic

  • FastAPI app factory in main.py using create_app().
  • Versioned API router from app/api/v1.
  • Environment-driven settings via app.config.settings.
  • Centralized exception handling returning JSON 500 errors.
  • Run locally with: uvicorn app.main:app --reload.

XML Flow Visualizer

Turns EIP-style integration XML — Apache Camel-like routes with sources, filters, routers, transforms and sinks — into a readable flow diagram. Ships as a double-click desktop app for macOS and Windows plus a CLI, and stays schema-agnostic: point it at any XML and it draws the tree.

  • Zero-dependency converter: XML → graph → Graphviz DOT using only the standard library.
  • Hybrid renderer: shells out to Graphviz for a polished hierarchical layout when the binary is installed; otherwise falls back to networkx + matplotlib, so it always works on a fresh machine.
  • Desktop app: Tkinter GUI with LR/TB layout switching, a DOT source tab, and PNG/SVG/DOT export.
  • Shipped, not just runnable: PyInstaller spec builds a standalone app that needs no Python installed.
Python Tkinter Graphviz matplotlib networkx PyInstaller pytest

  • Shape heuristics by element: from/to → ovals, choice/filter/when → diamonds, transform/map → boxes.
  • Edges labelled from uri / ref / condition / expression attributes.
  • One shared implementation in converter.py behind the CLI, the GUI and the tests.
  • Backend in use is reported in the app's status bar, so the fallback is visible rather than silent.
  • 9 pytest tests: namespace stripping, shape heuristics, graph counts, edge labels, DOT validity, quote escaping, malformed input.
  • Build per OS — PyInstaller cannot cross-compile, so the Mac app is built on macOS and the Windows exe on Windows.