Skip to content

PLAN.md

Latest commit

 

History

History
580 lines (463 loc) · 18.7 KB

PLAN.md

File metadata and controls

580 lines (463 loc) · 18.7 KB
Raw

ExamTutor: AI-Powered UK Exam Preparation Platform

Executive Summary

An AI-powered learning platform for UK students preparing for 11+ entrance exams and GCSEs, building on the architecture of the Ilya's Top 30 project. The platform will provide personalized tutoring, practice question generation, and curriculum-aligned learning using RAG over official educational content.

Market Analysis

Target Market Size

  • UK Private Tutoring Market: £7B annually
  • Global AI in Education: $3.43B (2023) → projected $54.5B (2032)
  • 11+ Candidates: ~100,000+ children annually across England
  • GCSE Students: ~700,000+ students per year cohort

Key Competitors

Platform Focus Pricing Strengths Gaps
Atom Learning 11+ only £49-99/month 90k users, 89% success rate, £25M Series A Stops at 11+, no GCSE
Seneca Learning KS2-A Level Free + Premium 2x faster learning claim, neuroscience-based Less personalized
Medly AI GCSE/A-Level 5% of tutor cost £1.7M funding, 10k users New, limited content
Third Space Learning Maths only School subscriptions AI tutor "Skye" Single subject
GoStudent Human tutors £20-40/hour Europe's largest Expensive, not AI-native

Our Differentiation

  1. Full spectrum: 11+ through GCSE (ages 9-16)
  2. RAG-powered: Deep understanding from curriculum documents, not just Q&A
  3. Open-source foundation: Built on proven DeepTutor architecture
  4. Knowledge graphs: LightRAG for interconnected topic understanding
  5. Mark scheme alignment: Teach exam technique, not just content

Exam Landscape

11+ Exams (Ages 10-11)

Providers

Provider Market Share Format Key Regions
GL Assessment ~70% Multiple choice, separate papers Kent, Bucks, Birmingham, Lincolnshire
CEM (Durham) ~25% Integrated papers, timed sections Berkshire, Bexley, Gloucestershire
ISEB ~5% Independent schools London independent schools
CSSE Regional Maths & English only Essex consortium

Subjects & Question Types

Verbal Reasoning (21 question types in GL)

  • Synonyms / Antonyms
  • Analogies (word relationships)
  • Code words (letter substitution, number codes)
  • Hidden words
  • Compound words
  • Letter sequences
  • Odd one out
  • Reading comprehension
  • Cloze passages (fill the gap)
  • Shuffled sentences

Non-Verbal Reasoning

  • Sequences and series
  • Matrices (grid completion)
  • Analogies (shape relationships)
  • Rotations and reflections
  • Transformations
  • Odd one out
  • Paper folding
  • Nets and cubes
  • Embedded shapes
  • Code breakers

Mathematics

  • National Curriculum KS2 content
  • Word problems
  • Fractions, decimals, percentages
  • Algebra basics
  • Geometry and measures
  • Data handling

English

  • Reading comprehension
  • Spelling and punctuation
  • Vocabulary
  • Grammar
  • Creative writing (some schools)

GCSEs (Ages 14-16)

Exam Boards

Board Market Share Known For Headquarters
AQA ~40% Largest, straightforward Manchester
Pearson Edexcel ~35% International presence London
OCR ~15% Cambridge-owned, analytical Cambridge
WJEC/Eduqas ~8% Wales + some England Cardiff
CCEA ~2% Northern Ireland Belfast

Core Subjects (Compulsory)

  • English Language (8700/1EN0)
  • English Literature (8702/1ET0)
  • Mathematics (8300/1MA1)
  • Combined Science or Triple Science

Popular Options

  • History, Geography
  • Modern Foreign Languages (French, Spanish, German)
  • Computer Science
  • Art & Design
  • Business Studies
  • Religious Studies
  • Physical Education

Grading

  • England: 9-1 scale (9 highest)
  • Wales/NI: A*-G scale
  • Grade 4 = "Standard pass" (old C)
  • Grade 5 = "Strong pass"

Content Sourcing Strategy

Tier 1: Official Free Sources (Priority)

Oak National Academy API ⭐ KEY RESOURCE

  • URL: https://open-api.thenational.academy/
  • License: Open Government License (free for commercial use)
  • Content: Lessons, videos, quizzes, transcripts for KS1-4
  • Integration: REST API for curriculum data
  • Coverage: All core subjects, aligned to National Curriculum

Government National Curriculum Documents

  • KS2 Framework: PDF
  • KS3/4 Framework: PDF
  • License: Crown Copyright, free to use

Exam Board Past Papers (Direct from boards)

Note: Exam boards allow educational use but redistribution requires care. Best approach:

  • Link to official sources rather than hosting
  • Use for RAG training (transformative use)
  • Generate similar-style questions rather than copying

GL Assessment Familiarisation Papers

  • Official sample papers available free
  • Purchase practice packs for fuller coverage

Tier 2: Open Educational Resources

BBC Bitesize

  • No API, but content is publicly accessible
  • 47% of UK students use it
  • Cannot redistribute, but can link/reference

Khan Academy

  • Creative Commons content
  • Good for foundational maths/science concepts
  • Less UK-curriculum aligned

Wikipedia / Simple Wikipedia

  • CC-BY-SA license
  • Background context for topics

Tier 3: Generated Content

AI-Generated Practice Questions

Using LLMs to generate:

  • Verbal reasoning questions (based on patterns)
  • Non-verbal reasoning (describe transformations)
  • Maths word problems
  • Comprehension passages with questions
  • Mark schemes

Quality Control Requirements:

  • Human review for accuracy
  • Cross-reference with curriculum objectives
  • A/B testing against real exam performance

Tier 4: Licensed/Partnership Content

Publisher Partnerships (Future)

  • CGP Books (dominant GCSE revision publisher)
  • Collins (11+ market leader)
  • Letts and Pearson revision guides
  • Exam board endorsed textbooks

School Partnerships

  • Access to anonymized performance data
  • Validation of question difficulty
  • Real-world testing

Technical Architecture

Core Stack (Inherited from AIEducator)

┌─────────────────────────────────────────────────────────┐
│                    Frontend (Next.js)                    │
│              React 19, TailwindCSS, port 3782           │
└─────────────────────────────────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────┐
│                   Backend (FastAPI)                      │
│                Python 3.12+, port 8001                   │
└─────────────────────────────────────────────────────────┘
                           │
           ┌───────────────┼───────────────┐
           ▼               ▼               ▼
    ┌────────────┐  ┌────────────┐  ┌────────────┐
    │   Agents   │  │  LightRAG  │  │  LLM API   │
    │ (Tutoring) │  │ (Knowledge)│  │ (OpenAI/   │
    │            │  │            │  │  Ollama)   │
    └────────────┘  └────────────┘  └────────────┘

New Components for ExamTutor

1. Question Bank System

class QuestionBank:
    - subject: str  # maths, english, vr, nvr
    - exam_type: str  # 11plus_gl, 11plus_cem, gcse_aqa, etc.
    - topic: str  # e.g., "fractions", "synonyms"
    - difficulty: int  # 1-5
    - question_text: str
    - answer_options: List[str]  # for multiple choice
    - correct_answer: str
    - mark_scheme: str  # how to award marks
    - worked_solution: str
    - source: str  # generated, oak_api, past_paper

2. Progress Tracking

class StudentProgress:
    - user_id: str
    - exam_target: str  # e.g., "gcse_maths_aqa"
    - topic_mastery: Dict[str, float]  # topic -> score 0-1
    - weak_areas: List[str]
    - practice_history: List[AttemptRecord]
    - predicted_grade: str

3. Adaptive Learning Engine

class AdaptiveLearner:
    def select_next_question(student: StudentProgress) -> Question:
        # Spaced repetition + weakness targeting

    def adjust_difficulty(performance: List[AttemptRecord]) -> int:
        # Dynamic difficulty adjustment

    def generate_revision_plan(target_date: date) -> Plan:
        # Countdown to exam scheduling

4. Mark Scheme Evaluator

class MarkSchemeEvaluator:
    def evaluate_answer(
        question: Question,
        student_answer: str,
        mark_scheme: MarkScheme
    ) -> EvaluationResult:
        # Use LLM to evaluate against mark scheme
        # Return: marks awarded, feedback, model answer

Data Models

Curriculum Structure

ExamBoard
├── Specification (e.g., AQA GCSE Maths 8300)
│   ├── Paper (Paper 1: Non-Calculator)
│   │   ├── Topic (Number)
│   │   │   ├── Subtopic (Fractions)
│   │   │   │   ├── LearningObjective
│   │   │   │   └── Questions[]

11+ Structure

ExamProvider (GL/CEM)
├── Subject (VR/NVR/Maths/English)
│   ├── QuestionType (e.g., "Code Words")
│   │   ├── Difficulty (1-5)
│   │   └── Questions[]

Agent Adaptations

From AIEducator to ExamTutor

AIEducator Agent ExamTutor Equivalent Adaptations
guide explain Age-appropriate language, curriculum links
solve work_through Show mark scheme alignment, exam technique
research explore_topic Link to Bitesize/Oak content
- practice NEW: Generate and evaluate practice questions
- revise NEW: Spaced repetition review
- mock_exam NEW: Timed exam simulation

New Agent: Practice Agent

class PracticeAgent:
    """Generate practice questions and evaluate answers"""

    def generate_question(
        subject: str,
        topic: str,
        difficulty: int,
        style: str  # "11plus_gl", "gcse_aqa"
    ) -> Question

    def evaluate_response(
        question: Question,
        response: str
    ) -> Feedback

    def explain_mistake(
        question: Question,
        wrong_answer: str,
        correct_answer: str
    ) -> Explanation

New Agent: Mock Exam Agent

class MockExamAgent:
    """Simulate timed exam conditions"""

    def generate_mock_paper(
        exam_type: str,
        duration_minutes: int
    ) -> MockPaper

    def run_timed_session(
        paper: MockPaper,
        student: Student
    ) -> ExamResult

    def generate_examiner_report(
        result: ExamResult
    ) -> Report

MVP Feature Set

Phase 1: Foundation (11+ Focus)

  • Oak Academy API integration
  • Question bank for VR/NVR/Maths/English
  • Basic practice mode with instant feedback
  • Progress tracking per topic
  • GL Assessment style questions

Phase 2: Intelligence

  • Adaptive difficulty adjustment
  • Weakness identification
  • Personalized revision plans
  • AI-generated questions
  • Parent progress reports

Phase 3: GCSE Expansion

  • AQA/Edexcel spec alignment
  • Past paper integration
  • Mark scheme evaluation
  • Subject-specific agents
  • Mock exam mode

Phase 4: Scale

  • Mobile app (React Native)
  • School/tutor dashboards
  • Performance analytics
  • Content partnerships
  • Gamification

Content Pipeline

Automated Ingestion

┌──────────────────┐     ┌──────────────────┐     ┌──────────────────┐
│   Oak API        │────▶│   ETL Pipeline   │────▶│   LightRAG       │
│   (Lessons)      │     │   (Transform)    │     │   (Index)        │
└──────────────────┘     └──────────────────┘     └──────────────────┘
         │                        │                        │
         ▼                        ▼                        ▼
┌──────────────────┐     ┌──────────────────┐     ┌──────────────────┐
│   Past Papers    │────▶│   Question       │────▶│   Question       │
│   (PDFs)         │     │   Extraction     │     │   Bank           │
└──────────────────┘     └──────────────────┘     └──────────────────┘

Question Generation Pipeline

# 1. Extract patterns from real questions
real_questions = load_past_paper("gl_vr_2023.pdf")
patterns = extract_patterns(real_questions)

# 2. Generate similar questions
generated = llm.generate(
    prompt=f"""Generate a {pattern.type} question
    similar in style to: {pattern.example}
    Topic: {pattern.topic}
    Difficulty: {pattern.difficulty}"""
)

# 3. Validate
validated = human_review(generated)

# 4. Store
question_bank.add(validated)

Legal Considerations

Safe to Use

  • Oak National Academy content (Open Government License)
  • National Curriculum documents (Crown Copyright)
  • Past papers for personal study/RAG training
  • AI-generated original questions

Requires Care

  • Past papers for redistribution (link, don't host)
  • Publisher content (licensing required)
  • BBC Bitesize content (reference, don't copy)

Must Avoid

  • Copying current year exam papers
  • Redistributing purchased practice papers
  • Using content without attribution where required

Recommended Approach

  1. Primary: Oak API + AI generation
  2. Secondary: Links to official exam board papers
  3. Future: Publisher licensing deals

Monetization Options

B2C (Direct to families)

  • Freemium: Basic practice free, advanced features paid
  • Subscription: £9.99-29.99/month (undercut Atom's £49-99)
  • Pay-per-mock: £2.99 per full mock exam

B2B (Schools/Tutors)

  • School licenses: £500-2000/year
  • Tutor tools: £29.99/month
  • White-label options

Marketplace

  • Tutor-created content (revenue share)
  • Premium question packs

Success Metrics

Learning Outcomes

  • Topic mastery improvement
  • Mock exam score progression
  • Correlation with real exam results

Engagement

  • Daily active users
  • Questions attempted per session
  • Session duration
  • Retention (week 1, month 1, month 3)

Business

  • Free → Paid conversion rate
  • Customer acquisition cost
  • Lifetime value
  • NPS score

Risks & Mitigations

Risk Impact Likelihood Mitigation
Content accuracy errors High Medium Human review, user reporting, rapid fixes
Copyright claims High Low Stick to OGL content, clear attribution
Competitor response Medium High Speed to market, differentiation
AI hallucination in answers High Medium Structured outputs, validation layers
Exam format changes Medium Medium Modular content system, quick updates

Implementation Roadmap

Month 1: Foundation

  • Fork AIEducator architecture
  • Oak Academy API integration
  • Basic question bank schema
  • 11+ VR question types (5 types)

Month 2: Core Features

  • All 21 VR question types
  • NVR question types
  • Practice mode with feedback
  • Basic progress tracking

Month 3: Intelligence

  • Adaptive difficulty
  • Weakness detection
  • AI question generation
  • Parent dashboard

Month 4: GCSE Start

  • GCSE Maths (AQA) spec mapping
  • Past paper integration
  • Mark scheme evaluation
  • Mock exam mode

Month 5-6: Polish & Launch

  • Mobile responsiveness
  • Performance optimization
  • Beta testing with families
  • Launch marketing

Next Steps

  1. Validate Oak API - Test integration, assess content coverage
  2. Design question schema - Finalize data models
  3. Build 11+ VR prototype - 5 question types, basic practice
  4. User testing - Find 10 families for feedback
  5. Iterate - Based on feedback, expand coverage

Appendix: Research Sources

Official Resources

Free Practice Resources

Competitor Research

Technical References