CASE STUDY / 03 · ACADEMIC WORKLOAD PLANNING

PACE / Make the workload realistic.

PACE (Planning Academic Capacity & Effort) is a student-focused workload planner. It uses task deadlines, estimated effort, priority and daily capacity to create a seven-day plan, then rebalances the schedule as work is completed.

Next.jsReactTypeScriptAppwriteVercel

01 / Problem statement

Why it exists.

A task list records obligations but does not answer whether the work can fit in the time available. PACE makes capacity and workload pressure part of the planning problem, helping surface a useful next task.

02 / System path

Architecture at a glance.

  1. NODE / 01Capture
  2. NODE / 02Understand
  3. NODE / 03Allocate
  4. NODE / 04Act
  5. NODE / 05Recalculate

STEP / 01

Capture assignments with subject/context, due date, estimated effort and priority.

STEP / 02

Use declared daily study capacity and a seven-day horizon to allocate the work.

STEP / 03

Expose workload pressure and explain why a task is recommended.

STEP / 04

Recalculate the plan after changes instead of treating the schedule as static.

03 / Engineering decisions

Trade-offs are part of the system.

DECISION / 01

Multi-user architecture

Authentication and storage use Appwrite. The documented design attaches an authenticated user ID to task rows and applies row-level permissions to protect user data.

DECISION / 02

Explainable planning

The product emphasizes deadlines, effort, priority and capacity instead of hiding the schedule behind an unexplained recommendation.

DECISION / 03

Disclose the foundation

PACE evolved from SYNAPSE. Its competition-specific interface, authentication, database architecture and workflow are described as an evolution—not as an unrelated invention.

04 / Failure modes & limits

Where it can fail.

05 / Source map

Where to inspect.

README.md

Documents Appwrite TablesDB, user isolation, seven-day planning and the relationship to SYNAPSE.

The map identifies only paths confirmed by available project documentation. Follow the repository link for the full current tree.

06 / External evidence