quiz-gen 0.6.1

AI-powered quiz generator for regulatory, certification, and educational documentation

quiz-gen

AI-powered quiz generator for regulatory documentation. Extract structured content from complex legal and technical documents to create comprehensive teaching and certification materials.

Features

• Multi-Agent Quiz Generation: Generate, validate, refine, and judge questions using configurable providers/models.
• EUR-Lex Document Parser: Parse and structure EU legal documents with full table of contents extraction
• Human-in-the-Loop: Integrate human input throughout the workflow.

Tech Stack

Backend

• Python — core package language
• FastAPI — serves the web UI and REST API from within the package
• LangGraph – multi-agent orchestration framework
• AI Providers: Anthropic (Claude), Cohere, Google (Gemini), Mistral, OpenAI
• SQLite – local database for documents and quizes organized by projects

Web UI

• React — interactive frontend
• Vite — fast dev server and production bundler (outputs to quiz_gen/ui/static)
• TypeScript — component and API code
• Tailwind CSS — utility-first styling

CLI

• argparse — flag-based CLI

Packaging

• PyPI — distributed as an installable Python package

Installation

pip install quiz-gen

Quick Start

Interactive UI

Parse documents and generate quiz questions in an integrated responsive UI:

quiz-gen --ui

The UI lets you go from a raw document to a finished quiz without writing any code. You paste a EUR-Lex URL or upload an HTML file, click Generate TOC, and immediately see the full document structure in a navigable table of contents. Click any article to load its parsed content, optionally edit it inline to focus the AI on a specific passage, then click Generate Quiz to run the five-agent pipeline right there in the browser. Each agent's output — generator drafts, validator scores, refiner edits, and the judge's final decision — are displayed in collapsible sections so you can inspect exactly how the questions were produced and catch any issues before using them. This is faster for exploration and quality review than running scripts, because there is no round-trip to the terminal and no JSON to read manually.

Multi-Agent Quiz Generation

Quiz generation uses four specialized agents (conceptual, practical, validator, refiner, and judge). Providers are configurable per agent, with supported providers: Anthropic, Cohere, Google, Mistral, and OpenAI. Any text-generation model name from these providers can be passed directly. The package relies on provider defaults for generation parameters.

Multi-Agent Architecture and Configuration

from quiz_gen.agents.workflow import QuizGenerationWorkflow
from quiz_gen.agents.config import AgentConfig

config = AgentConfig(
    conceptual_provider="cohere",
    conceptual_model="command-a-03-2025",
    practical_provider="google",
    practical_model="gemini-3-pro-preview",
    validator_provider="openai",
    validator_model="gpt-5.2-2025-12-11",
    refiner_provider="anthropic",
    refiner_model="claude-sonnet-4-5-20250929",
    judge_provider="mistral",
    judge_model="mistral-large-latest",
)

workflow = QuizGenerationWorkflow(config)
result = workflow.run(chunk)

Parsing EUR-Lex Documents

from quiz_gen import EURLexParser

# Parse a regulation document
url = "https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202401689"
parser = EURLexParser(url=url)
chunks, toc = parser.parse()

# Access structured content
print(f"Extracted {len(chunks)} content chunks")
print(f"Document has {len(toc['sections'])} major sections")

# Save results
parser.save_chunks('output_chunks.json')
parser.save_toc('output_toc.json')

Working with Chunks

# Iterate through extracted chunks
for chunk in chunks:
    print(f"{chunk.title}")
    print(f"Type: {chunk.section_type.value}")
    print(f"Number: {chunk.number}")
    print(f"Content: {chunk.content[:200]}...")
    print(f"Hierarchy: {' > '.join(chunk.hierarchy_path)}")
    print()

Displaying Table of Contents

# Print formatted TOC
parser.print_toc()

# Output:
# PREAMBLE
#   Citation 
#   Recital 1
#   Recital 2
#   ...
# 
# ENACTING TERMS
#   CHAPTER I - PRINCIPLES
#     Article 1 - Subject matter and objectives
#     Article 2 - Scope

Development

Setting up Development Environment

# Clone the repository
git clone https://github.com/yauheniya-ai/quiz-gen.git
cd quiz-gen

# Install with development dependencies
pip install -e ".[dev]"

# Run tests
pytest --cov=src --cov-report=term-missing

# Run linting
ruff check .
black .

Project Structure

quiz-gen/
├── data/                          # Local data files
│   ├── raw/                       # Source HTML documents
│   ├── processed/                 # Parsed chunks and TOC JSON
│   └── quizzes/                   # Generated quiz output
├── docs/                          # MkDocs documentation source
├── examples/                      # Runnable example scripts
│   ├── eur_lex_html_file.py
│   ├── eur_lex_html_url.py
│   └── quiz_gen_multi_model.py
├── src/
│   └── quiz_gen/                  # Package source
│       ├── agents/                # Multi-agent system
│       │   ├── config.py          # AgentConfig dataclass
│       │   ├── conceptual_generator.py
│       │   ├── practical_generator.py
│       │   ├── validator.py
│       │   ├── refiner.py
│       │   ├── judge.py
│       │   └── workflow.py        # LangGraph orchestration
│       ├── parsers/
│       │   └── html/
│       │       └── eur_lex_parser.py
│       ├── ui/                    # FastAPI + React static bundle
│       │   ├── server.py
│       │   ├── api.py
│       │   └── static/
│       ├── utils/
│       │   └── helpers.py
│       └── cli.py
├── tests/
│   ├── test_agents/
│   ├── test_cli/
│   ├── test_parsers/
│   └── test_utils/
├── pyproject.toml
├── README.md
├── CHANGELOG.md
└── .env

API Reference

AgentConfig

Dataclass that configures every agent in the multi-agent pipeline. API keys and base URLs are loaded automatically from environment variables when not provided directly.

Provider / model settings (per agent – defaults shown):

Parameter	Default provider	Default model
conceptual_provider / conceptual_model	openai	gpt-4o
practical_provider / practical_model	anthropic	claude-sonnet-4-20250514
validator_provider / validator_model	openai	gpt-4o
refiner_provider / refiner_model	openai	gpt-4o
judge_provider / judge_model	anthropic	claude-sonnet-4-20250514

Supported provider values: openai, anthropic, google, mistral, cohere.

Workflow settings:

• auto_accept_valid: bool = False — skip judge when validation already passes
• save_intermediate_results: bool = True
• output_directory: str = "data/quizzes"
• min_validation_score: int = 6 — minimum score (out of 10) to pass validation
• strict_validation: bool = True
• max_retries: int = 3
• verbose: bool = True

Methods:

• validate() — raises ValueError if config is invalid
• save(filepath, verbose=False) — write config to JSON
• load(filepath) (classmethod) — load config from JSON
• print_summary() — print a human-readable config table

QuizGenerationWorkflow

LangGraph-based orchestration of the five-agent pipeline.

from quiz_gen.agents.workflow import QuizGenerationWorkflow
from quiz_gen.agents.config import AgentConfig

config = AgentConfig()          # reads API keys from environment
workflow = QuizGenerationWorkflow(config)

# Single chunk
result = workflow.run(chunk)

# Batch
results = workflow.run_batch(chunks, save_output=True, output_dir="data/quizzes")

Methods:

• run(chunk, improvement_feedback=None) → Dict — run the full pipeline for one chunk; returns full state including final_questions, judge_decision, validation_results, and errors
• run_batch(chunks, save_output=True, output_dir="data/quizzes") → List[Dict] — run for multiple chunks, optionally saving each result to JSON

Individual Agents

Agents can be used standalone outside of the workflow:

from quiz_gen.agents.conceptual_generator import ConceptualGenerator
from quiz_gen.agents.practical_generator import PracticalGenerator
from quiz_gen.agents.validator import Validator
from quiz_gen.agents.refiner import Refiner
from quiz_gen.agents.judge import Judge

Class	Key method	Returns
ConceptualGenerator	generate(chunk, improvement_feedback=None)	Dict question
PracticalGenerator	generate(chunk, improvement_feedback=None)	Dict question
Validator	validate(qa, chunk) / validate_batch(qas, chunk)	Dict / List[Dict]
Refiner	refine(qa, validation_result, chunk) / refine_batch(qas, validation_results, chunk)	Dict / List[Dict]
Judge	judge(conceptual_qa, practical_qa, chunk)	Dict with decision and reasoning

---

EURLexParser

Main parser class for EUR-Lex documents.

Methods:

• parse() -> tuple[List[RegulationChunk], Dict]: Parse document and return chunks and TOC
• fetch() -> str: Fetch HTML content from URL
• save_chunks(filepath: str): Save chunks to JSON file
• save_toc(filepath: str): Save table of contents to JSON file
• print_toc(): Display formatted table of contents

RegulationChunk

Represents a parsed content chunk (article or recital).

Attributes:

• section_type: Type of section (ARTICLE, RECITAL, etc.)
• number: Section number (e.g., "1", "42")
• title: Full title including subtitle
• content: Text content
• hierarchy_path: List of parent sections
• metadata: Additional structured data

SectionType

Enumeration of document section types.

Values:

• PREAMBLE: Preamble section
• ENACTING_TERMS: Main regulatory content
• CITATION: Citation in preamble
• RECITAL: Recital in preamble
• CHAPTER: Chapter division
• SECTION: Section within chapter
• ARTICLE: Article (main content unit)
• ANNEX: Annex section

CLI

The quiz-gen command provides two independent modes: document parsing and web UI.

Document parsing

# Parse from URL and save chunks + TOC JSON
quiz-gen https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32018R1139

# Parse a local HTML file
quiz-gen data/raw/regulation.html

# Print the table of contents without saving any files
quiz-gen --print-toc --no-save regulation.html

# Save output to a custom directory with verbose logging
quiz-gen --verbose --output results/ regulation.html

Option	Description	Default
INPUT	URL or local HTML path	required
-o, --output DIR	Output directory for JSON files	data/processed
--chunks FILENAME	Custom filename for chunks JSON	<id>_chunks.json
--toc FILENAME	Custom filename for TOC JSON	<id>_toc.json
--no-save	Parse and display statistics without writing files	—
--print-toc	Print formatted table of contents to console	—
--verbose	Show detailed progress and error stack traces	—
-v, --version	Print version and exit	—

Web UI

# Launch UI on http://localhost:8000 and open browser automatically
quiz-gen --ui

# Custom host and port
quiz-gen --ui --host 127.0.0.1 --port 9000

# Launch without opening a browser tab
quiz-gen --ui --no-browser

# Development mode with auto-reload and debug logging
quiz-gen --ui --reload --log-level debug

Option	Description	Default
--ui	Start the FastAPI/uvicorn server	—
--host HOST	Server bind address	0.0.0.0
--port PORT	Server port	8000
--reload	Auto-reload on code changes (development)	—
--no-browser	Do not open a browser tab on start	—
--log-level LEVEL	Uvicorn log level (debug/info/warning/error)	warning

Use Cases

Compliance and Legal

• Analyze regulatory requirements systematically
• Support automated document analysis workflows
• Build searchable knowledge bases from legal texts

Education and Training

• Generate study materials from regulatory documents
• Create structured learning paths for certification programs
• Extract key concepts for examination preparation

Supported Document Types

Currently supports:

• EUR-Lex HTML Documents: European Union regulations, directives, decisions

Document Format Requirements

• Documents must use EUR-Lex HTML format
• Must contain eli-subdivision elements for proper structure identification
• Supports multi-level hierarchies with chapters, sections, and articles

TODOs

• Save results by project in a local database
• Integrate human feedback
• Support for additional document formats (PDF, DOCX, PPTX)
• Multi-language support for UI
• Light/Dark scheme for UI

License

This project is licensed under the MIT License. See the LICENSE file for details.

Support

• Documentation: https://quiz-gen.readthedocs.io
• Issue Tracker: https://github.com/yauheniya-ai/quiz-gen/issues

Contributing

Contributions are welcome! Please ensure:

1. Code follows PEP 8 style guidelines
2. All tests pass: pytest --cov=src --cov-report=term-missing
3. New features include appropriate tests
4. Documentation is updated