Solving AI Workflow Issues for Content Teams

Background

Our learning platform's content team creates quiz-based learning material (MCQs) for students. To speed up content creation, they used AI tools like ChatGPT, but their workflow had many problems that made the process inefficient and mentally draining.

The Problem

Even though AI was part of the process, it was still inefficient due to several workflow issues:

• Outputs were often not aligned with the syllabus
• Many repeated or similar questions generated
• Needed manual fact-checking and verification
• Depended on multiple tools: ChatGPT, Excel, Google, Docs
• High mental effort to manage the process end-to-end
• Prompts had to be manually crafted each time
• Users constantly switched between tools
• Accuracy of every question had to be checked manually

My Role & Initiative

The Goal

I wanted to build one single tool where the content team could:

• Internal Prompt System - Questions generated from subject + topic + source with controlled outputs
• One-Tap Review Flow - Save or discard questions with single tap, flagged items sent to manual review
• Swipe-to-Verify - Swipe gesture opens inline Google results for quick fact-checking
• UI for Long Sessions - Soft color palette optimized for extended review sessions
• Subject & Syllabus Creation - Structured content organization system
• Source Material Integration - Add books, text, and materials for AI reference
• Auto-Difficulty Assignment - Automatic difficulty level assignment by grade
• Duplicate Prevention - Eliminate duplication of generated questions

Key Features & Design

Key Screens & Features

Signup & Onboarding

Clean and intuitive signup process to get content teams started quickly.

Home Dashboard (Empty State)

Welcome screen for new users with clear guidance on getting started.

Home Dashboard (With Content)

Main dashboard showing subjects, recent activity, and quick access to key features.

Subject Detail Page

Detailed view of a subject with syllabus structure and management options.

Subject Configuration

Setup and configuration interface for creating new subjects with structured syllabuses.

Pre-Generation Confirmation

Confirmation screen before generating questions, allowing users to review settings and parameters.

Topic Selection from Syllabus

Interactive syllabus browser for selecting specific topics to generate questions from.

Loading Screen (Pre-Generation)

Engaging loading state while AI processes and generates questions based on selected parameters.

Question Sorting Page

Interface for organizing and categorizing generated questions by difficulty, topic, or other criteria.

Content Validation

One-tap review interface for validating generated content with save, discard, and flag actions.

Saved Questions

Repository of all saved questions with search, filter, and management capabilities.

Add New Subject

Form interface for creating new subjects with syllabus structure and source material integration.

AI Quality Assurance System

To ensure the highest quality of AI-generated educational content, I developed a comprehensive prompt engineering system with multiple layers of quality control and validation.

System Prompt Architecture

The AI system uses a two-part prompt strategy with carefully crafted system prompts that establish the AI as an expert educational content creator.

Content Creation Expertise

• Clear, unambiguous question writing
• Grade-appropriate language adaptation
• Effective distractor design based on common misconceptions

Assessment Standards

• Alignment with educational objectives
• Question validity and reliability
• Balanced difficulty distribution

Database Design & Planning

Before building the application, I carefully planned the database structure with handwritten notes and relationship diagrams.

Dynamic Context Prompting

The main prompt is dynamically constructed based on provided context and includes specific input parameters for precise control.

Topic & Subtopic

Primary subject area and specific focus

Difficulty & Grade Level

Customized complexity and age-appropriateness

Source Content

Educational material for question basis

Question Count

Precise number of questions to generate

Quality Control Requirements

Each question must meet strict design standards to ensure educational value and consistency.

1

Uniqueness

Each question tests different aspects, avoids duplication

2

Structure

Clear language, scenario-based problems, critical thinking

3

Answer Options

4 options, plausible distractors, consistent format

4

Explanations

Comprehensive analysis of correct and incorrect answers

Output Format & Validation

The AI must return a valid JSON array with strict formatting requirements for reliable parsing and processing.

{
  "text": "Question text goes here?",
  "type": "multiple_choice",
  "options": ["Option A", "Option B", "Option C", "Option D"],
  "correctAnswer": "Option A",
  "explanation": {
    "correct": "Why Option A is correct",
    "distractors": {
      "Option B": "Why Option B is incorrect",
      "Option C": "Why Option C is incorrect",
      "Option D": "Why Option D is incorrect"
    },
    "commonMisconceptions": ["Misconception 1", "Misconception 2"],
    "relatedConcepts": ["Concept 1", "Concept 2"]
  },
  "difficulty": "medium",
  "cognitiveLevel": "knowledge",
  "topic": "Topic Name",
  "subtopic": "Subtopic Name"
}

Format Validation

Strict JSON parsing with error handling

Content Validation

Required fields and data type checking

Results & Impact

What I Learned

• - Taking initiative on internal problems creates real impact
• - Learned the basics of backend, APIs, and database design as a designer
• - Debugging taught me how to unblock quickly with AI + developer input
• - Better prompt logic directly improves AI output quality
• - Simple UX makes repetitive tasks easier and less tiring
• - Vibe coding gave me confidence to build and ship my own MVP