This is the hardest thing I've built. Dilly is the daily workbench that keeps college students ahead of AI, not a chatbot bolted onto a landing page. It's live on the App Store: a living profile, one clear move a day, a resume ledger that only uses real facts, a 50K-job feed refreshed nightly, and Stripe billing. I founded it, designed it, and wrote the code end to end, over 1,100 production builds and counting. Every skill below had to work at 2 a.m. when a sophomore actually opens the app.
Product architecture
I designed three products that share one spine (auth, metering, AI, and profile facts) so career, admissions, and medical don't become three separate codebases. The skill was knowing what to share and what to isolate before the complexity ate me.
Expo / React Native
I shipped the client in Expo Router: push notifications, on-device speech-to-text, calendar and contacts, share-card capture, Reanimated motion. Native enough to feel like a real app; fast enough that I can still ship weekly from a dorm.
FastAPI + PostgreSQL
The backend is a large FastAPI surface on AWS RDS: resume parsing, ATS scoring, internship crawl ingest, cron jobs, SSE streaming. I treat the API as the product: if the route isn't trustworthy, the AI is just theater.
LLM systems
Claude and GPT dual-route depending on the job. Embeddings match students to roles. A voice agent detects intents and runs tools. Memory extraction, interview feedback, and output safety keep the coach useful without inventing a student's life. The skill is the system around the model.
Auth & payments
Email OTP, magic links, Sign in with Apple and Google, Stripe subscriptions with webhook idempotency, Moves-based metering, even Apple Wallet passes. Monetization and identity aren't afterthoughts. They're product features I own.
Ops & infra
Docker on Railway, Sentry when things break, GitHub Actions for crawlers and health, a Next.js recruiter dashboard, Resend for mail. Founder engineering means the night shift is also mine.