RAG in the Wild: A Case Study

A case study implementing and comparing advanced RAG pipelines (RAG Fusion, HyDE, CRAG, and Graph RAG) on the CRAG dataset, with evaluation metrics and an interactive web demo.

Tech Stack

• Python (retrieval, pipelines, evaluation, Flask API)
• React + Vite (frontend)
• FAISS + sentence-transformers (global corpus retrieval)

Pipelines

• RAG Fusion
• HyDE
• CRAG
• Graph RAG

Quick Start

1. Install dependencies:

pip install -r requirements.txt
cd frontend
npm install
cd ..

2. Configure:

• Copy config/config.example.yaml to config/config.yaml
• Set dataset/model values (dataset_path, generation_model, top_k)
• Use Groq/Gemini or another free/local provider (no OpenAI key)

3. Run backend:

python backend/app.py

4. Run frontend:

cd frontend
npm run dev

5. Run evaluation:

python run_evaluation.py

Evaluation output is saved to results/evaluation_results.csv.

Dataset

• CRAG Task 1 & 2 dev dataset
• Place crag_task_1_and_2_dev_v4.jsonl in dataset/
• Schema: docs/dataset.md

---

Report

Full implementation notes, pipeline analysis, and evaluation results:

View Report →

Pipeline extension notes (Basic RAG, Multi-Query RAG, RRR integration):

View Pipeline Addendum →

Notes

• Full assignment details: ASSIGNMENT.md
• Keep folder structure unchanged