Nōritsu

This is a phone app that used ai to create content for personal motivation and curated the content to fit into the user's schedule. So the app can motivate users and also to reminds them what they should do next, kinda a nice way to assist users' schedule and keep everything aligned.

🧩 Tech Stack Frontend: I used React (with Vite) for a fast, lightweight, and responsive UI, tailored for both mobile and desktop. The interface focuses on minimalist inspiration — clean typography, smooth animations, and a feed-style layout that combines daily AI prompts and real founder stories.

Backend/API: The backend is built using Node.js + Express, connected to a PostgreSQL database via Prisma ORM. It handles user auth, progress tracking, and serves curated content (AI-generated and human-written) via a RESTful API.

Authentication: I integrated Google OAuth via Firebase Auth for smooth onboarding and session handling.

AI Integration:

I use OpenAI’s GPT-4 API to generate daily motivational messages, micro-journaling prompts, and personalized affirmations based on user behavior and mood logs.

For more immersive content, I also fine-tuned GPT-3.5 on a small corpus of 1,000+ startup and resilience-themed founder stories using OpenAI’s embedding + retrieval setup — this creates a mini story engine that surfaces highly relevant narratives.

Storage & Delivery:

Supabase Storage for user-uploaded voice journals and story bookmarks.

Cloudflare Pages and Vercel Edge Functions for blazing fast delivery and occasional dynamic server-side rendering.

Mobile Optimization: Though it's a PWA for now, I used Tailwind CSS and Framer Motion to deliver a delightful experience on mobile — where most users engage during their morning or evening routines.

⚙️ Hacky and Notable Bits I used a "mood-to-content" mapping system — a lightweight rules engine powered by embeddings and user sentiment logging. It tailors content not just by what users like, but also how they feel that day. This led to higher engagement and retention.

Founder stories are semi-structured in Markdown with embedded tags, then parsed and ranked using cosine similarity via OpenAI embeddings + a Pinecone vector DB. This gives the app a “Spotify Discover Weekly” feel — but for motivation.

To get early test data, I ran a prompt-powered crawler that scraped hundreds of first-person founder interviews from blogs and podcasts, then chunked and tagged them using GPT before feeding them into the system.