04 · Projects

Things I've built.

2026

2 projects
Full-Stack · Volunteer Project · Sole Developer · April 2026

GMSM-USA: Community Platform for GMSM America

Production community platform for GMSM America, a nationwide US religious and cultural organization of several hundred families that previously coordinated through spreadsheets and WhatsApp threads. Built solo as volunteer work for the community; live since April 2026 and in daily use. Built around one constraint: most members have no email address, so authentication is a custom phone-number credential layer (bcrypt-hashed in the browser, stored against a synthetic email so Supabase Auth, the credentials table, and the directory record share one UUID) with Google OAuth and account identity linking alongside it. Families register as households rather than individuals — 79 households cover 293 people, 86% of all member records, behind 136 login accounts. Authorization is enforced in PostgreSQL, not the client: 47 row-level security policies, 20 security-definer RPCs, three-tier RBAC driven by a unit-tested permission matrix, and database triggers that reject writes to closed events. Signup collapsed from three sequential client writes into a single atomic Postgres function after a failure mid-flow left orphaned auth users unrecoverable. Real-time updates run over Supabase WebSocket channels backed by Postgres logical replication, gated on auth state. Two-tier caching (in-memory session cache with in-flight request deduplication over a localStorage stale-while-revalidate layer) clears on sign-out for shared family computers. 89 kB gzipped initial JS via route-level code splitting across 12 pages. 340 active members across 45 states and 93 cities; 463 registrations across 17 events including a 14-city national tour. 78 files, ~26k lines of frontend, ~3.1k lines of SQL, 43 migrations, 9 tables, 68 unit tests.

React 19ViteReact RouterSupabasePostgreSQLPL/pgSQLRow-Level SecuritybcryptRead the case study →
Full-Stack · Academic Project · May 2026

Next-Step: Full-Stack School Management Platform

Full-stack school management platform built as a 5-person academic project. Async REST API with FastAPI exposing 20+ endpoints across 10 router modules backed by 9 SQLAlchemy ORM models (cascade deletes, JSON columns, unique constraints), with Alembic migrations on PostgreSQL and auto-generated Swagger / ReDoc docs. JWT auth with Argon2 hashing and Google OAuth 2.0 PKCE flow drives role-aware access for Admin, Teacher, and Student via factory-pattern middleware (get_current_user, require_admin, require_roles). Cross-platform React Native + Expo app (iOS, Android, Web) with TypeScript, Zustand state management, file-based routing via Expo Router, and AsyncStorage session persistence. Google Drive API integration for document management, Celery + Redis with Flower for background task monitoring. Co-developed via feature-branch Git workflow with 11+ pull requests, pytest + httpx async integration tests, and AI-assisted development tooling.

FastAPIPostgreSQLSQLAlchemyAlembicCeleryRedisFlowerJWT

2024

2 projects
AI/ML Core · Game AI & RL · Academic Project · November 2024

TrackMania Reinforcement Learning Agent

Autonomous racing agent in PyTorch implementing Implicit Quantile Networks (IQN) with NumPy-based state processing, reaching 85% track completion through iterative reward shaping. Training loop optimized with CUDA-accelerated parallel environment simulations, cutting training time by 60%.

PythonPyTorchIQNReinforcement LearningDeep RLCUDANumPydxcamGitHub →
AI/ML Core · Creative AI & Image Processing

Interactive Image Mosaic Generator

Built a Gradio app that generates artistic image mosaics using vectorized NumPy pipelines, giving users a fast, interactive UI to control mosaic density and style.

PythonNumPyGradioImage ProcessingComputer VisionVectorized OperationsGitHub →Live demo →

2023

1 project
Data Science · Social Media Analytics

Twitter Sentiment Analysis (NLP Project)

Built an end-to-end sentiment analysis pipeline over Twitter data, combining API ingestion, cleaning, feature extraction, and classical ML classifiers for topic-level sentiment.

PythonTwitter APINLPMachine LearningPandasNLTKTextBlobGoogle ColabGitHub →
Reach out

Got a question? Let's talk.