I built an app for a teacher to save her time grading.

2026-05-31 · Source: LLM on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, long

Summary

An app named CorregIA was developed to significantly reduce the time a Language Arts teacher in Chile spends grading 70 open-ended tests. The application transcribes paper tests, suggests scores and comments using the teacher's specific rubric, and allows for human validation. Initially, a simpler script automated document assembly and spelling checks, saving three hours, but the core grading burden remained. CorregIA addresses this by enabling teachers to upload their rubrics and example answers for AI calibration. It presents AI-generated scores and comments alongside test photos and transcriptions, allowing the teacher to accept or edit, and prioritizes doubtful cases for review. Key design choices include using the teacher's own OpenRouter API key for cost and model flexibility, ensuring human validation for all grades, and respecting the teacher's question-by-question grading workflow. The app also integrates macOS Continuity Camera for efficient paper test capture and provides post-grading analytics.

Key takeaway

For AI Engineers or Product Managers building custom tools, prioritize deeply understanding a specific user's workflow and pain points. This project demonstrates that augmenting human decision-making with AI proposals, rather than full automation, can yield significant time savings and user adoption. Focus on integrating existing user habits and providing transparent AI reasoning, ensuring the human remains in control for critical validation. Consider local data storage and user-provided API keys to minimize infrastructure costs and enhance model flexibility.

Key insights

AI tools can effectively augment human expertise by proposing drafts for validation, significantly reducing repetitive tasks.

Principles

• AI should propose, human should decide.
• Tailor AI to user's specific criteria.
• Design for existing user workflows.

Method

Design an app by first defining the "what" and "why" (PRD), then the "how" (architecture, data model), and a roadmap, before coding with AI assistance guided by fixed rules and specialized agents.

In practice

• Use Continuity Camera for document scanning.
• Encrypt API keys in OS keychain.
• Order AI proposals by confidence.

Topics

• AI-assisted Grading
• Custom Application Development
• Human-in-the-Loop AI
• Educational Technology
• Workflow Automation
• OpenRouter API

Best for: AI Engineer, Software Engineer, AI Product Manager

Related on AIssential

• Not Everything Should Be Automated: The Line We’re Forgetting to Draw with AI — Thoughtful AI integration augments human judgment for critical tasks, rather than replacing it entirely.
• What are Examples of AI Applications in the Classroom? — AI offers personalized learning, adaptive content delivery, and early disability detection in education.
• Building My Own Personal AI Assistant: A Chronicle, Part 2 — API-driven data access and AI-powered task decomposition significantly enhance personal agent efficiency and utility.
• Collaborative AI Systems: Human-AI Teaming Workflows — Effective human-AI collaboration combines AI's pattern recognition with human judgment for superior outcomes.
• How AI Is Reshaping UX Design — And How I’m Learning to Work With It — AI accelerates UX design workflows by reducing friction between ideas and execution, amplifying human judgment.
• Domain expertise still wanted: the latest trends in AI-assisted knowledge for developers — Developers increasingly use AI for learning and work, but a significant trust gap necessitates validation with traditional resources.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by LLM on Medium.