edurag 0.1.3

An LLM-based Retrieval-Augmented Generation (RAG) component for education, supporting problem solving, academic literature analysis, and knowledge Q&A.

EduRAG

An LLM-based Retrieval-Augmented Generation (RAG) component for education, supporting problem solving, academic literature analysis, and knowledge Q&A.

Features

• Two RAG Modes: SimpleRAG (simple & efficient) and AgenticRAG (intelligent reasoning)
• Multi-LLM Support: OpenAI GPT series, Google Gemini, Ollama local models
• Multiple Document Formats: PDF, DOCX, DOC, TXT, Markdown
• Teacher Persona Customization: Custom teacher name, subject, grade level, teaching style
• Multi-turn Conversation: Context-aware continuous Q&A
• Vector Store Persistence: Avoid repeated embedding, fast knowledge base loading

Installation

pip install edurag

Install optional dependencies:

# Use AgenticRAG (based on LangGraph)
pip install edurag[agentic]

# Use Google Gemini
pip install edurag[gemini]

# Use Ollama local models
pip install edurag[ollama]

# Install all optional dependencies
pip install edurag[all]

Quick Start

Basic Usage

from edurag import SimpleRAG

# Initialize
rag = SimpleRAG(api_key="your-openai-api-key")

# Load documents
rag.load_documents("textbook.pdf")

# Ask questions
answer = rag.ask("What is the main content of this document?")
print(answer)

Custom Teacher Persona

from edurag import SimpleRAG, TeacherProfile

# Create teacher profile
teacher = TeacherProfile(
    name="Mr. Wang",
    subject="High School Physics",
    grade_level="Grade 12",
    teaching_style="Focus on concept understanding, good at explaining abstract principles with real-life examples",
    introduction="20 years of teaching experience, physics competition coach"
)

# Initialize RAG
rag = SimpleRAG(
    api_key="your-openai-api-key",
    teacher_profile=teacher
)

# Load textbooks
rag.load_documents([
    "physics_chapter1.pdf",
    "mechanics_topics.docx"
])

# Ask - AI will respond as Mr. Wang with his teaching style
answer = rag.ask("Why is the acceleration of free fall constant?")

Using Different LLMs

# OpenAI
rag = SimpleRAG(
    api_key="sk-xxx",
    llm_provider="openai",
    llm_model="gpt-4o"
)

# Google Gemini
rag = SimpleRAG(
    api_key="your-google-key",
    llm_provider="gemini",
    llm_model="gemini-pro"
)

# Ollama local model (no API Key required)
rag = SimpleRAG(
    llm_provider="ollama",
    llm_model="llama3"
)

Persistent Vector Store

# First time: auto-save vector store
rag = SimpleRAG(
    api_key="sk-xxx",
    vectorstore_path="./my_knowledge_base"
)
rag.load_documents("./documents/")

# Later: load existing store (skip embedding)
rag = SimpleRAG.from_vectorstore(
    vectorstore_path="./my_knowledge_base",
    api_key="sk-xxx"
)

AgenticRAG (Intelligent Reasoning)

Suitable for complex problems. The Agent autonomously decides whether to retrieve and supports multi-step reasoning.

from edurag import AgenticRAG, TeacherProfile

teacher = TeacherProfile(
    name="Mr. Wang",
    subject="High School Physics",
    grade_level="Grade 12",
    teaching_style="Good at analogies, explains with real-life examples"
)

rag = AgenticRAG(
    api_key="sk-xxx",
    teacher_profile=teacher
)

rag.load_documents(["physics_textbook.pdf"])

# Agent automatically decides whether to retrieve, supports multi-step reasoning
answer = rag.ask("Compare the similarities and differences of Newton's three laws")

# View reasoning process
result = rag.ask_with_steps("Explain the law of conservation of momentum")
print(result["steps"])  # View Agent's reasoning steps
print(result["answer"]) # Final answer

SimpleRAG vs AgenticRAG:

Feature	SimpleRAG	AgenticRAG
Retrieval	Always retrieves	Agent decides
Reasoning	Single-step	Multi-step
Speed	Faster	Slower
Cost	Lower	Higher
Use Case	Simple Q&A	Complex analysis

API Reference

SimpleRAG

Main RAG class providing document loading and Q&A functionality.

Initialization Parameters

Parameter	Type	Default	Description
api_key	str	None	LLM API key
llm_provider	str	"openai"	LLM provider: openai/gemini/ollama
llm_model	str	"gpt-4o"	Model name
teacher_profile	TeacherProfile	None	Teacher persona configuration
config	EduRAGConfig	None	Full configuration object

Methods

• load_documents(sources): Load documents into knowledge base
• ask(question): Ask and get answer
• ask_with_sources(question): Ask and return answer with source documents
• search(query, top_k): Directly search relevant documents
• clear_history(): Clear conversation history
• save_vectorstore(path): Save vector store

TeacherProfile

Teacher persona configuration class.

from edurag import TeacherProfile

teacher = TeacherProfile(
    name="Mr. Li",              # Teacher name
    subject="Mathematics",       # Teaching subject
    grade_level="Middle School", # Grade level
    teaching_style="...",        # Teaching style
    introduction="...",          # Introduction (optional)
    language="English"           # Response language
)

EduRAGConfig

Full configuration class for advanced customization.

from edurag import EduRAGConfig

config = EduRAGConfig(
    llm_provider="openai",
    llm_model="gpt-4o",
    api_key="sk-xxx",
    temperature=0.7,           # Generation temperature
    chunk_size=1000,           # Document chunk size
    chunk_overlap=200,         # Chunk overlap
    retrieval_top_k=4,         # Number of retrieval results
    vectorstore_path=None      # Vector store path
)

Preset Teacher Templates

from edurag.prompt.teacher_profile import PRESET_TEACHERS

# Available presets
teacher = PRESET_TEACHERS["physics_senior"]    # High School Physics
teacher = PRESET_TEACHERS["math_college"]      # College Mathematics
teacher = PRESET_TEACHERS["english_junior"]    # Middle School English
teacher = PRESET_TEACHERS["chemistry_senior"]  # High School Chemistry

License

MIT