Group scheduling app that finds a time that works for everyone across their Google and Outlook calendars.

• ▸FastAPI backend with a provider abstraction over Google Calendar API and Microsoft Graph (MSAL), syncing busy slots across multiple OAuth-connected accounts and writing finalized events to each participant's own calendar.
• ▸Deterministic scheduling algorithm that ranks candidate windows (incl. overnight) by conflict count and preferred-slot overlap to surface the top 5 meeting times.
• ▸Building an agentic chat assistant (LangGraph + Gemini) that creates and edits events through conversation and books only after user confirmation, covered by 160 backend and frontend tests.
• ▸PostgreSQL schema with Row Level Security and a trigger that enforces one write calendar per user; live RSVP updates via Supabase Realtime and hourly calendar syncs via APScheduler.

Automated job search pipeline that finds, scores, and tracks applications so you only see jobs worth applying to.

• ▸Pipeline that scrapes LinkedIn and HiringCafe listings, scores each against a parsed resume with Gemini, and writes only 80+/100 matches to Google Sheets with deduplication.
• ▸Daily GitHub Actions job that reads Gmail, classifies application replies (rejection, assessment, interview, offer) with rules plus a Gemini fallback, updates each application's status, and sends Discord alerts.
• ▸Companion tracker that polls Google, Amazon, Netflix, Apple, Microsoft, and Meta career portals via their JSON APIs on a GitHub Actions cron and posts new matches to Discord.

AI health assistant that triages messages by severity and runs every answer through a safety check before replying.

• ▸LangGraph agent pipeline that triages each message by severity (emergency, clinical, general, off-topic), injects the user's health profile, vitals, and medications as context, and routes drafts through a critic node that sends unsafe responses back for up to 2 revisions.
• ▸Eval harness of 52 labeled cases gating prompt/model changes on accuracy and emergency recall, validating a move to a lighter Gemini triage model at 97% accuracy, 7/7 emergency recall, and 0 over-escalations.
• ▸Co-developed the FastAPI backend with a partner, streaming responses over Server-Sent Events while a parallel summarizer node titles new conversations.

iOS app for Bay Area climbers: see how busy each bouldering gym is and which one to hit right now.

• ▸Building a native SwiftUI iOS app backed by a FastAPI service and a hosted Supabase (Postgres + PostGIS) database covering 16 Bay Area gyms.
• ▸Designed a 10-table schema with PostGIS geography columns for location-based recommendations, versioned with Alembic migrations alongside Supabase CLI extension migrations.
• ▸Secured the API by verifying Supabase ES256 JWTs against the project's JWKS endpoint; CI runs ruff, strict mypy, pytest, and xcodebuild tests on every push.
• ▸In progress: Playwright-based ingestion of live crowd levels and a ranking that picks the best gym for your location right now.