Pacing

Students don’t burn out because they can’t do the work — they burn out because they can’t see it coming. Pacing is a mobile app that makes academic load visible at three time scales at once, and surfaces opportunities only when there’s genuine capacity for them.

The design came out of two user-research briefs rather than a hunch. The findings that shaped it: stress is a perceived gap between demands and resources, not an absolute workload; first-year students plan at a horizon of one to two days, not a semester; external accountability is the missing ingredient in every tool they’d tried; and setup friction kills adoption before any feature matters. Positioned against Notion, Todoist, Google Calendar, Flourish, and Woebot, the proactive-academic quadrant was empty.

What it does

The app opens on a Day view — a 24-hour timeline showing real calendar events as solid blocks alongside dashed “ghost blocks” proposing work that fits the gaps: deadline prep, opportunities, or a walk based on recent check-ins. Tap a ghost block to accept it and it lands on your Google Calendar; tap empty space to block your own time. Pinch out to a Week strip, out again to a Semester heatmap of colour-coded load — three zoom levels onto the same calendar.

Around that sit an Advisor that scans Gmail and the web for opportunities and surfaces them as suggestions when they fit a real slot, a commitment gate that answers “should I take this on?” with a week-by-week load projection, and a low-friction daily check-in parsed by an LLM to extract friction signals.

Engineering

React Native (Expo) on the front end, FastAPI and PostgreSQL on the back, Celery and Redis for the work that can’t happen in a request cycle — iCal polling, Calendar sync, nightly load recomputation, Gmail history processing. Claude handles the language-shaped parts: check-in parsing, opportunity extraction, and the commitment-gate reasoning. There’s a separate Vite/React web prototype deployed alongside the mobile app.

A few decisions worth calling out. Course deadlines come in over iCal rather than the Moodle REST API, because a student can export their own calendar URL without involving university IT — the REST path needs admin permission the user doesn’t have. The Gmail scanner reads only subject, from, and a 100-character snippet; the message body is never fetched or stored. Weekly load lives in its own table rather than being aggregated per request, which keeps the heatmap a single read instead of a join across thousands of events.

The interaction design followed the research too. Suggestions are ghost blocks pinned to specific time slots rather than another to-do list, because a list is just one more thing to triage — tying each suggestion to a slot you actually have preserves agency while removing the planning burden that the interviews identified as the main reason people abandoned their previous tool. And navigation is a durable bottom tab bar with pinch-to-zoom layered on top, not pinch alone: users predicted they’d abandon the app, so discoverability had to beat delight.

Repository: github.com/abhineet-agarwal/pacing