Skip to main content

Comprehensive benchmark and evaluation framework for educational AI question generation

Project description

InceptBench

PyPI version Python Version License Swagger

Educational content evaluation framework using LLM-based analysis.

WebsiteBenchmarksAPI EndpointAPI DocsGitHub

Overview

InceptBench evaluates educational content across multiple quality dimensions:

  • Automated Classification - Determines content type (question, quiz, reading, etc.)
  • Hierarchical Evaluation - Decomposes complex content and evaluates bottom-up
  • Comprehensive Metrics - 8-11 metrics per content type with scores and reasoning
  • Curriculum-Aware - Integrates curriculum standards via vector store search

Installation

pip install inceptbench

System Dependencies

InceptBench requires system-level Cairo libraries for inline SVG image analysis. Without these, SVG images in educational content will not be analyzed.

macOS:

brew install cairo pango gdk-pixbuf libffi

Ubuntu/Debian:

sudo apt-get install -y libcairo2 libpango-1.0-0 libpangocairo-1.0-0 libgdk-pixbuf2.0-0 libffi-dev

Windows: Follow the GTK3 installer for Windows or install via MSYS2:

pacman -S mingw-w64-x86_64-cairo mingw-w64-x86_64-pango mingw-w64-x86_64-gdk-pixbuf2

CLI Usage

# Create sample input file
inceptbench example

# Evaluate from JSON file
inceptbench evaluate content.json

# Evaluate raw content
inceptbench evaluate --raw "What is 2+2? A) 3 B) 4 C) 5 D) 6"

# Save results to file
inceptbench evaluate content.json -o results.json

# Check version
inceptbench --version

REST API

The production API is available at https://api.inceptbench.com

# Health check
curl https://api.inceptbench.com/health

# Interactive docs
# Visit: https://api.inceptbench.com/docs

# Evaluate content
curl -X POST https://api.inceptbench.com/evaluate \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer your-api-key" \
  -d '{"generated_content": [{"content": "What is 2+2? A) 3 B) 4 C) 5 D) 6"}]}'

Programmatic Usage (Python)

import asyncio
from inceptbench import EvaluationService

async def main():
    service = EvaluationService()
    result = await service.evaluate(
        content="What is 2+2? A) 3 B) 4 C) 5 D) 6",
        curriculum="common_core"
    )
    print(f"Score: {result.overall.score:.2f}")

asyncio.run(main())

Input Format

{
  "generated_content": [
    {
      "id": "q1",
      "curriculum": "common_core",
      "request": {
        "grade": "7",
        "subject": "mathematics",
        "type": "mcq",
        "difficulty": "medium",
        "locale": "en-US",
        "skills": {
          "lesson_title": "Congruent and Similar Triangles",
          "substandard_id": "CCSS.MATH.CONTENT.7.G.A.1+3"
        },
        "instruction": "A real-world problem involving congruent and similar triangles"
      },
      "content": "Triangle ABC is similar to triangle DEF. If AB = 6 cm and DE = 9 cm, what is the ratio of their corresponding sides?"
    }
  ]
}

Content Item Fields

Field Required Default Description
content Yes - Content to evaluate (string or JSON)
id No Auto-generated Unique identifier
curriculum No common_core Curriculum for alignment
request No null Generation metadata (see below)

Request Metadata Fields (all optional)

Field Description
grade Grade level (e.g., "7", "K", "12")
subject Subject area (e.g., "mathematics", "english")
type Content type (e.g., "mcq", "fill-in", "article")
difficulty Difficulty level (e.g., "easy", "medium", "hard")
locale Locale/language code (e.g., "en-US", "es-MX")
skills Skills info (JSON object or string)
instruction Generation instruction/prompt

Images

Images are automatically detected from content. Include as:

  • Direct URLs: https://example.com/image.png
  • Markdown: ![alt](https://example.com/image.png)
  • HTML: <img src="https://example.com/image.png">
  • Inline SVG: <svg>...</svg> (requires Cairo system libraries, see System Dependencies)

Content Types

The evaluator automatically classifies content into:

Type Description
question Single educational question
quiz Multiple questions together
fiction_reading Fictional narrative passages
nonfiction_reading Informational passages
other General educational content

Documentation

For complete documentation, input format details, and developer guides:

View Full Documentation on GitHub

License

Proprietary - Copyright Trilogy Education Services

Project details


Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

inceptbench-2.5.8.tar.gz (9.7 MB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

inceptbench-2.5.8-py3-none-any.whl (9.8 MB view details)

Uploaded Python 3

File details

Details for the file inceptbench-2.5.8.tar.gz.

File metadata

  • Download URL: inceptbench-2.5.8.tar.gz
  • Upload date:
  • Size: 9.7 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: poetry/2.4.1 CPython/3.12.13 Linux/6.6.87.2-microsoft-standard-WSL2

File hashes

Hashes for inceptbench-2.5.8.tar.gz
Algorithm Hash digest
SHA256 2d79a3e3f1e345c5ff31bb25c591d96a8047afc0ee5678ebdde2cce90a5d9bbe
MD5 464ce8bf62daa4f5bed62d2bf6f3f847
BLAKE2b-256 48cc7b530acbbdf077d5ab879411f7d6094cac748b09ae9dc2a75845c76d4364

See more details on using hashes here.

File details

Details for the file inceptbench-2.5.8-py3-none-any.whl.

File metadata

  • Download URL: inceptbench-2.5.8-py3-none-any.whl
  • Upload date:
  • Size: 9.8 MB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: poetry/2.4.1 CPython/3.12.13 Linux/6.6.87.2-microsoft-standard-WSL2

File hashes

Hashes for inceptbench-2.5.8-py3-none-any.whl
Algorithm Hash digest
SHA256 5ca01e508b26163699e032705ebb1961091af1c3df9c69fe2800060e850c2983
MD5 22bc929b010c652311904e06a0298993
BLAKE2b-256 5cee67b88f5c4dc7c107c4f13220eba2408edfc748ac9193e18421db0231484d

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page