[Projects]

Built & Deployed

From field infrastructure to embedded systems and AI prototypes, I make systems designed to work in production.

Flagship Project

Deployed 2026 · Ongoing

MMDM · Offline-First Clinical Platform

Medical Missionaries of Divine Mercy · Houston → Costa Rica · 2024–Present

Purpose-built offline clinical platform for a six-clinic volunteer medical mission across a four-building parish campus in Costa Rica. Started as an OpenEMR rebuild on Linux; evolved into a custom Node.js / SQLite / React system designed from first principles around the mission's actual constraints: no internet, non-technical volunteers, and hundreds of patients in a single week.

Node.jsSQLiteReactLinux (Kubuntu)systemdTailwind CSSLAN-onlyOffline-firstRBAC
Read case study

Other Projects

air.grid - Live U.S. Air Quality & Emissions

Live 2026

Geospatial / Data

Live atlas of what's in the air across the U.S.: industrial emitters sized by emissions, real-time air quality sensors, and the schools and neighborhoods next to them, together on one interactive map. 268,980 facilities, 15,897 live sensors, real proximity joins, no mock data.

  • Built and deployed in one day (Handshake AI Projects challenge, June 1, 2026) using Claude Code's multi-agent system: parallel ingestion subagents coordinated through a frozen data contract, with sequential gates for schema, joins, and QA.
  • Pulls from 6 federal sources (EPA ECHO, FRS, TRI, GHGRP, NCES, Census ACS) plus AirNow, PurpleAir, and OpenAQ sensor networks; a geo-matcher joins 108,336 schools to their nearest emitters with scipy cKDTree.
  • deck.gl over MapLibre renders 10k+ points at 60fps, with independent toggle layers for facilities, air quality, schools, demographics, and NWS wind drift.
  • Hourly scheduled job keeps sensor readings live; every figure shows its timestamp and source, and an Analysis view ranks regions and surfaces the most-exposed campuses.
Next.js 14TypeScriptTailwindMapLibre GLdeck.glRechartsPythonscipy cKDTreeVercel

Atlas · Cinematic 3D Travel Globe

Live 2026

3D / Geospatial

Interactive 3D globe pinning everywhere I've lived, vacationed, and done mission work, plus everywhere I plan to go. Styled after the Apple lock-screen Earth: deep space, city lights, atmospheric scattering, soft bloom. Custom-built with three.js and globe.gl plus hand-written GLSL; no UI kits, no state libraries.

  • Custom day/night shader pipeline: normal-mapped terrain, soft terminator blend, isolated and boosted city lights, Blinn-Phong ocean glint, and rotating cloud shadows cast onto the surface.
  • Camera-relative sun direction recomputed every frame so the lighting reads cinematic from any angle; the day/night toggle sweeps the terminator across the globe in about a second.
  • Procedural starfield of 4,500 stars from a seeded PRNG, identical on every load, with per-star size and color tiers for a realistic luminosity distribution.
  • Automatic performance mode detects software rendering and low-RAM devices before any WebGL allocation, then falls back to lite shaders, no MSAA, and halved geometry for roughly 50x less fill work.
  • 107-node append-only place tree validated at build time; URL-hash deep links fly the camera straight to any pin.
Next.js 14TypeScriptthree.jsglobe.glCustom GLSLTailwind CSSStatic export

Emergency Evacuation Guidance Device

Prototype

Embedded · IoT

ENGR 1201 · Houston City College · Spring 2026

Handheld emergency evacuation device that computes and displays real-time escape routes from live temperature sensor data. Routes rebuild automatically as heat develops; reroutes around blocked zones, escalating through warning states, and triggering Shelter-In-Place when all exits are blocked.

  • 5 DS18B20 probes on a shared 1-Wire bus monitor 10 building zones (6 rooms, 2 hallways, 2 exits), streaming one CSV line per second to the Pi over USB-serial at 115,200 baud.
  • Dijkstra's algorithm via NetworkX rebuilds the optimal evacuation route in under 1 ms on every sensor state change; edge weights are dynamically scaled by zone temperature.
  • D-pad navigation drives a top-down Pygame floor plan; RGB LED and piezo buzzer mirror the active threat state in real time.
Raspberry Pi 4ESP32Python · PygameNetworkXUSB-SerialSystemd

foreCastor

Private

Finance · Personal

A PIN-locked personal finance tracker with live bank integration, spending analytics, and a long-term savings goal engine, built entirely for personal use.

  • Connected to the bank via the Plaid API to automatically import, normalize, and categorize transactions without manual entry.
  • Tracks a long-term savings goal with a compound-growth projection chart and a target date that changes color based on whether you're on pace.
  • Per-category monthly spending limits with live progress bars that shift from teal to amber to red as you approach and exceed your budget.
  • Scans transaction history to detect recurring charges and suggests adding them to your fixed expenses list automatically.
  • 31 unlockable achievements tied to real savings milestones, income streaks, and usage habits, with rarity tiers styled after Steam.
Next.js 15TypeScriptTailwind CSSCloudflare PagesSQLitePlaid API

More on GitHub

Coursework, lab repos, automation scripts, and works-in-progress.

github.com/osakhra →