Pastel
Built a journaling platform that allows users to record memories and schedule them for future delivery using Next.js, Clerk, and PostgreSQL.
The Problem
Journaling apps要么太复杂要么太简单. Most are just text editors with a calendar view. I wanted something that felt personal, a space where writing to your future self had weight, and where your emotional patterns surfaced over time.
Architecture
Pastel is a Next.js application with a PostgreSQL backend. Clerk handles authentication, and the AI summary feature uses an LLM API to analyze entry themes and emotional arcs.
The data model is straightforward: users create entries (text + mood tag + optional photo), which are organized into chronological "letters" that can be scheduled for future delivery.
Key Decisions
Clerk Auth was chosen for speed of implementation. Building auth from scratch is a rite of passage, but for this project the focus was on the product experience, not the infrastructure.
Server Components first. Most of the app renders on the server. Client-side JavaScript is reserved for interactive elements like the entry editor, mood picker, and real-time summary generation.
AI summaries analyze a user's entries over a configurable window (week, month, quarter) and surface patterns: recurring themes, emotional trends, notable moments. The prompt engineering here was iterative; getting the AI to be insightful without being presumptuous was the hardest part.
What I Learned
The most interesting challenge was designing for emotional weight. A journal entry isn't a tweet. It needs space, privacy, and a sense of permanence. Every design decision (typography, spacing, animation) was in service of that feeling.