## Competition Overview
Welcome to the Retrieval-Augmented Generation (RAG) Challenge! This competition tests your ability to build an intelligent system that can accurately answer questions about textbook content while providing precise references to support its answers.


### Dataset License
This competition uses the OpenStax Psychology (2e) textbook, licensed under Creative Commons Attribution 4.0 International License (CC BY 4.0). Access for free at openstax.org. Source: https://openstax.org/details/books/psychology-2e

## Description
### Problem Statement
In this competition, you'll work with a psychology textbook dataset. Your task is to:
1. Build a RAG system that can accurately answer questions about the textbook content
2. Provide precise references to relevant sections and page numbers that support each answer
3. Ensure answers are both accurate and faithful to the source material

### Challenge Goals
- Demonstrate effective information retrieval from structured text
- Generate accurate, context-aware answers
- Provide verifiable references to source content
- Balance answer quality with retrieval precision


## Dataset
### What's Provided
- Queries.json file containing questions
- Test set with questions (without answers)
- Full textbook content in structured format
- Sample submission file showing expected format

### For Submission format, refer to dataset description

## Model Constraints
Models must be open source and have ≤3 billion parameters
Allowed models include (but not limited to):
Llama-2 3B
FLAN-T5 3B
MPT 3B
TinyLlama 1.1B
Phi-2 2.7B

### Vector Database
The baseline implementation uses Pinecone as the vector database, but you're free to use any vector database that supports similarity search, including:
- Chroma
- Weaviate
- Qdrant
- Milvus
- FAISS
- MongoDB with vector search

The choice of vector database should support:
- Efficient similarity search
- Storage of embeddings and metadata
- Fast query response times
- Easy integration with Python

Your performance may vary based on vector database selection and configuration. Ensure your chosen solution can handle the textbook embeddings within competition resource constraints.

The evaluation metric remains the same regardless of vector database choice.

## Rules and Requirements
1. Only use publicly available, free models and resources
2. No external API calls during inference
3. Submissions must include all required columns in the specified format
4. Solutions must be reproducible and well-documented

## Resources
- Sample code for evaluation metrics
- Baseline RAG implementation
- Submission format guide
- Discussion forum for technical questions