Projects.

A multi-room IoT monitoring system that turns a Raspberry Pi into a touchscreen hub for the home. Wi-Fi-enabled ESP32 sensor nodes capture environmental data including temperature, humidity, light, and ambient audio, then stream it via HTTP/JSON to a central Flask server. The dashboard aggregates live readings across rooms alongside integrated productivity apps. Built end-to-end from sensor firmware through API to UI.

• 3 sensor types per node (DHT22, BH1750, MAX4466) interfaced over I2C and GPIO.
• 6 integrated apps including weather, timers, and system stats on the touchscreen UI.
• JSON API endpoints aggregating real-time readings from every node into one dashboard.

A real-time multi-sport dashboard streaming live scores and play-by-play across MLB, NHL, NBA, and NFL. A league adapter pattern normalizes data from each sport's REST API into a common schema, while WebSocket push delivers updates to the browser the moment they hit the server. Built with sport-specific win probability models and an in-game alert engine that surfaces upsets, comebacks, and momentum shifts as they happen.

• 4 win probability models tuned per sport: 9,072-state Markov lookup (MLB), Poisson-probit (NHL), Stern probit (NBA), 15-feature logistic regression (NFL).
• 4 league adapters normalizing REST API feeds into a unified play-by-play stream.
• Alert engine detecting upsets, comebacks, and momentum shifts during live games.

A constant-current synchronous Buck LED driver designed to power high-intensity turn signals on Midnight Sun's solar race vehicle. Built around the automotive-grade MPM6010 power module with a low-side current-sensing feedback topology to hold luminosity steady across the vehicle's full battery range. The 2-layer Altium board prioritizes EMC and thermal performance, with PWM control routed through the Vehicle Control Unit so blinking patterns are software-driven instead of relying on mechanical relays.

• 10V to 16V input range with constant-current feedback holding luminosity stable across battery sag.
• 2-layer Altium PCB with input caps placed adjacent to the IC to shrink the high-frequency switching loop.
• Wide copper pours and AEC-Q100 qualified components for sustained operation under high thermal and vibration stress.

An early-warning wrist device for delirium risk in hospitalized patients. The wearable captures motion and environmental data through an Arduino-based sensor stack, with custom firmware handling activity tracking and sleep-environment monitoring. A Flask-SocketIO backend runs real-time ML inference in PyTorch to classify activity, then streams live readings and inactivity alerts to a React dashboard. Built end-to-end from sensor firmware through model inference to UI.

• 3-axis accelerometer, thermistor, and I2C LCD on an Arduino sensing core running custom firmware.
• 95%+ activity classification accuracy from a PyTorch ML model trained on labeled motion data.
• React dashboard surfacing live readings and nurse-facing inactivity alerts.

A compact 10V to 15V to 3.3V step-down buck regulator on a 2-layer PCB, designed end-to-end in Altium Designer at 2A output. The power stage was sized directly from the IC datasheet, with duty cycle, inductor ripple, output capacitor selection, and feedback compensation all calculated to hit the 3.3V rail target. Layout work focused on a tight hot-loop, solid ground plane, and clean decoupling, with full DRC/ERC verification, BOM, and Gerbers ready for fabrication.

• 10V to 15V input regulated down to a 3.3V, 2A output on a 2-layer compact PCB.
• Power stage sized from the IC datasheet: duty cycle, inductor ripple, output caps, and feedback compensation.
• Tight hot-loop layout with wide current paths, solid ground plane, and labeled test pads, verified DRC/ERC clean.

A Python CLI that pulls market data for BTC, ETH, and AAPL from Yahoo Finance (with an offline demo mode), then cleans and aligns OHLCV bars at 5-minute or daily resolution for scanning. A configurable rule engine evaluates technical indicators including SMA/EMA crossovers, RSI, and MACD with adjustable lookbacks and warm-up periods, firing alerts and structured logs when thresholds or crossovers trigger. Includes a lightweight backtester with cost and slippage modeling to sanity-check rules on historical data.

• 3 assets (BTC, ETH, AAPL) on 5-minute and daily OHLCV bars from Yahoo Finance with offline demo fallback.
• 4 indicators (SMA, EMA, RSI, MACD) feeding a rule engine with cool-downs, configurable lookbacks, and structured alert logs.
• Backtester reporting PnL, win rate, and max drawdown with cost and slippage modeling.

A series of VRC competition robots built from scratch across three seasons of Spin Up, Over Under, and High Stakes. Each robot is a full-stack mechanical and software build: custom drivetrains and scoring mechanisms designed in CAD, paired with autonomous and driver-control code in C++ on the V5 platform. PID-tuned routines and match data feed back into mechanical revisions every cycle, with a complete engineering notebook capturing the design decisions and iterations behind each build.

• World-class competition robots across 3 seasons: 1x Excellence Award, 3x Tournament Champions.
• PID-tuned autonomous routines refined through practice-field testing and match data review.
• Full engineering notebook documenting design decisions, iterations, and post-competition redesigns.