Backend systems, finance tooling, and a few academic experiments. Source is on GitHub where the repo is public.
Integration hub bridging LINE Shop, the LINE Messaging API, and Supabase for a merit-offering delivery service. Staff upload fulfillment proof, admins approve, and customers get automatic LINE push notifications. Daily PDF and CSV reports, plus a test mode that reroutes bot messages to debug safely in production.
Near real-time social platform. Live video views, reactions, comments, push notifications. Load-balanced WebSocket connections, persistent storage for offline notification replay.
Streamlit app projecting future portfolio value via Monte Carlo. Pulls historical prices from Yahoo Finance, calibrates drift and a covariance matrix from log returns, then runs multivariate Geometric Brownian Motion paths with Cholesky decomposition to preserve asset correlations. Vectorized NumPy engine, interactive Plotly distributions.
Personal net-worth tracker — Next.js frontend, FastAPI backend over Supabase Postgres, wired as a Turborepo monorepo. Logs monthly balances across cash, Bitcoin, and a US equity portfolio, pulls live prices and USD/THB FX via yfinance, and charts holdings over time. Portfolio rebalancing against stored target allocations.
Student attendance tracker on Rails 8. Each student carries a generated UID rendered as a QR code; scanning records a timestamped entry, with a Turbo-powered feed updating live as check-ins arrive. Custom bcrypt session auth, Ransack search, Pagy pagination, soft-deletes via Discard.
Forward- and reverse-mode automatic differentiation in Scala 3. Expressions modeled as a sealed hierarchy of case classes — sums, products, powers, trig — each carrying its own differentiation rules, parsed from an interactive prompt via parser combinators. Derivatives run sequentially and in parallel with Futures, timings reported side by side.
Instruction-set simulator for the MIPS R-series architecture, in C. Decodes each 32-bit word into opcode and register fields, then dispatches R-, I-, and J-type instructions to update register and memory state cycle by cycle. Shell for loading programs, stepping execution, and dumping state for verification.
Numerical simulation of a chaotic double pendulum. Derives the equations of motion from the Lagrangian, solves the coupled accelerations each step with Newton-Raphson over the Jacobian, and integrates the trajectory with scipy odeint. Renders the motion as a matplotlib animation.
Full-stack rental app with separate customer and admin sides. Search models by availability dates, add-ons, and a booking-to-payment flow; admins manage cars, bookings, customers, and reviews. Availability computed with date-range SQL joins and a stored procedure that assigns a free car from the chosen model.
Notebook analysis of a marketing-campaign customer dataset. pandas cleaning, seaborn and matplotlib exploratory analysis, then a scikit-learn decision tree — GridSearchCV tuning, train/test split, confusion matrix — to predict campaign response.