EduMindAI-backend

EduMindAI-backend is an advanced Retrieval-Augmented Generation (RAG) backend designed to generate exams and questions that faithfully match official exam topics, grade levels, difficulty, and structure. Leveraging a multi-agent architecture, the system ensures that every generated exam mirrors the template, style, and rigor of real, official exams.

Objective

• Generate exams and questions that are indistinguishable from official exams—right topics, right grade, right difficulty, right structure.
• When a user requests an exam, the system produces questions and compiles the exam using the same template and structure as the official exams.

How It Works

1. Official Exam Ingestion & Indexing

• Ingests a large collection of official exam PDFs and text files.
• Parsing: Extracts text from each exam (using Azure Form Recognizer/OpenAI Vision for PDFs, or direct text loading).
• Chunking: Splits parsed text into overlapping, manageable chunks for fine-grained retrieval.
• Embedding & Metadata Extraction: Each chunk is embedded using OpenAI Embeddings and tagged with metadata (subject, branch, title, grade, topic) using an LLM.

2. Semantic Retrieval

• User requests (e.g., "Generate a grade 10 math exam on algebra and geometry") are parsed to extract subject, grade, and topics.
• The request is embedded and used to retrieve the most relevant chunks from the official exam database, filtered by metadata to ensure the right grade and topic.

3. Exam Generation & Structure Matching (Multi-Agent Orchestration)

• Exam Analysis Agent:

  • Analyzes the user’s request and the structure of official exams.
  • Splits the exam into several exercises, each with its own topic, grade, and question type, matching the official template.

• Parallel Exercise Generation Agents:

  • For each exercise, a dedicated agent generates a new question or exercise.
  • Each agent uses the retrieved official exam context to ensure the generated content matches the required topic, grade, difficulty, and detail.
  • All exercises are generated in parallel for efficiency.

• Formatting Agent:

  • After all exercises are generated, a formatting agent assembles them into a single, well-structured exam document.
  • The output matches the official exam template (sections, headings, formatting, etc.).

• Clarification Agent:

  • If the user’s request is missing key information (grade, topic, format), a clarification agent asks follow-up questions to ensure accuracy.

4. User Experience

• Input: User submits a request (e.g., "Generate a grade 10 math exam on algebra and geometry").
• Clarification: If needed, the system asks for missing details.
• Output: The system returns a complete exam, with each exercise matching the structure, style, and difficulty of official exams.

---

Quickstart

Prerequisites

• Python 3.9+
• pip

Installation

# Clone the repository
 git clone <repo-url>
 cd EduMindAI-backend

# Install dependencies
 pip install -r api/requirements.txt

Configuration

Edit api/config.yaml to set your API keys and model parameters:

• openai_api_key: Your OpenAI API key
• azure_formrecognizer_key and azure_formrecognizer_endpoint: For Azure Form Recognizer (optional)
• exams_path: Path to the folder containing official exam documents

Running the API

uvicorn api.main:app --reload

The API will be available at http://127.0.0.1:8000/.

Main Endpoints

• POST /api/clarify — Checks if the user request is clear or needs more info
• POST /api/chat — Generates an exam or questions based on the user’s request

---

Technical Architecture

System Overview

EduMindAI-backend is built around a multi-agent RAG architecture:

• ExamDataPipeline: Ingests, parses, chunks, and embeds official exam documents.
• VectorStore: Stores and retrieves document chunks using semantic similarity and metadata.
• ExamQuestionAgent: Orchestrates retrieval, context assembly, and multi-agent exam generation.
• API Endpoints: Expose chat, search, and clarification functionality via FastAPI.

Multi-Agent Exam Generation Flow

flowchart TD
    A["User"] -->|"POST /api/clarify"| B["Clarification Agent"]
    B -->|"If clarification needed"| A
    B -->|"If clear"| C["POST /api/chat"]
    C --> D["ExamQuestionAgent"]
    D --> E["Retrieval Pipeline"]
    E --> F["PDF Parsing"]
    F --> G["Chunking"]
    E --> H["Embedding"]
    D --> I["Exam Analysis Agent"]
    I --> J["Parallel Exercise Generation Agents"]
    J --> K["Formatting Agent"]
    K --> L["Response: Official Exam Template"]
    L --> A
    G --> H
    H --> E
    E -->|"Context"| D

Loading

---

Development

• Main app: api/main.py
• Chatbot logic: api/chatbot.py
• Exam data pipeline: api/rag/exam_data_pipeline.py
• Requirements: api/requirements.txt

Linting & Formatting

You may use tools like black and flake8 for code quality.

Troubleshooting

If you encounter the following error:

Your system has an unsupported version of sqlite3. Chroma requires sqlite3 >= 3.35.0

You can solve it by commenting the lines from 85 to 103 in the file YOUR_PATH/chromadb/__init__.py.

License

This project is licensed under the MIT License.