Problem

College work isn’t hard because the material is hard — it’s hard because it’s scattered. Deadlines live in PDFs, half of what matters is only said out loud in lecture, and a general chatbot has never seen your syllabus. After one messy semester at GSU, I wanted something that kept me connected to the course itself, not another tab of notes.

What I built

An iPhone app (plus web and Chrome sync) where you upload each syllabus once. Built as founder at CourseConnect AI LLC; in beta with Main Street Foundry, free on TestFlight.

• one timeline of classes and deadlines
• lecture recording that becomes notes and catches spoken due dates
• a tutor per class that knows your syllabus
• grades imported from Safari (D2L/Canvas) without your school password
• lock-screen widgets and calendar sync

What I learned

The “connect classmates” chat idea was the wrong half of the name — hard to get right, and not what people needed first. The syllabus-on-your-phone pivot was. Also: if a professor only says a deadline out loud, recording isn’t a nice-to-have — it’s the product. I still use it every day; most of what’s in the app came from living in it, not from a roadmap.

Problem

My mom was diagnosed with Parkinson’s, and freezing of gait can stop someone mid-step with little warning. Clinic care doesn’t cover the hallway at home — I wanted something on her wrist that could notice a freeze and help the next step come.

What I built

An iPhone + Apple Watch cueing aid (not a medical device) that learns your walk, runs on the Watch without the phone nearby, and taps a gentle rhythm when steps stall. Built for my mom, and for anyone else who deals with this. Pilot open at gaitguard-app.vercel.app.

• 30-second calibration that sets personal freeze thresholds and cue tempo
• Watch-first rhythmic haptics (60–130 BPM) when a freeze hits
• thumbs up/down after each cue so sensitivity adapts to you
• optional beat-while-walking metronome, steadiness score, history/trends + CSV export
• private by default — data stays on device, no account

What I learned

HealthKit + WatchConnectivity force you to design for the wrist first. False positives hurt trust faster than a missed cue — and building for someone you love makes the honesty bar higher: say what it is, what it isn’t, and invite real feedback early.

Problem

Group chats die in “where should we go?” loops. By the time someone picks a place, the night’s gone. We wanted plans that feel scarce and real — venues you can actually walk into.

What I built

An open-source 12-hour Hacklanta build: AI proposes outings for your city with real Google Places data, then crews claim them over Supabase Realtime before someone else does.

• AI plan pool for your area
• live claim / release with scarcity
• real venues, photos, and maps
• shareable crew briefing page

What I learned

Scarcity makes people decide. Real Places data beats fake cards. Shipping in 12 hours means cutting every feature that isn’t the claim loop.

Problem

Camera rolls bury the trip. I wanted a place that treats photos as a route — pins on a map, string between stops — not another infinite grid.

What I built

A photo board for the world: geotag shots, connect them with red string, and play the trip back. Built with help from my friend Steven after I got into photography.

• map pins from EXIF / place
• red string between stops
• play-the-route camera fly
• public profiles and stories

What I learned

The map is the product. Private-by-default matters. And once you can play a trip, a flat gallery feels dead — which is why this portfolio reuses the same idea.