paramjeetn/ragbench-backend repository overview

⁠RagBench Backend

FastAPI-based evaluation engine and API service for RagBench — an open-source platform for evaluating, comparing, and optimizing RAG (Retrieval-Augmented Generation) pipelines.

⁠🚀 Quick Start with Docker Compose

The easiest way to run the complete RagBench stack is with the pre-configured Docker Compose file.

⁠1. Download the Docker Compose file

curl -O https://raw.githubusercontent.com/paramjeetn/RagBench/main/docker-compose.hub.yml

⁠2. Start the stack

docker compose -f docker-compose.hub.yml up -d

⁠3. Access the services

Once the containers are running:

To stop the stack:

docker compose -f docker-compose.hub.yml down

⁠🛠️ Features

• Document Ingestion — Parse PDF, Markdown, and TXT documents, chunk them, generate embeddings, and store them in Qdrant.

• Hybrid Retrieval — Dense vector search, sparse BM25 retrieval, and Hybrid Search using Reciprocal Rank Fusion (RRF).

• Reranking — Cross-encoder reranking using ms-marco-MiniLM-L-12-v2.

• RAG Evaluation — Evaluate pipelines using:

  • Faithfulness
  • Answer Relevancy
  • Contextual Precision
  • Contextual Recall

• Multiple LLM Providers — Gemini, OpenAI, Anthropic, and Ollama.

⁠🏗️ Architecture

RagBench consists of:

Frontend
   │
   ▼
FastAPI Backend
   │
   ├── Document Ingestion
   ├── Retrieval
   ├── Reranking
   ├── RAG Evaluation
   │
   ├── Qdrant
   └── LLM Providers

The Docker Compose setup provides the required services and allows the stack to be started with a single command.

⁠🔗 Links

• GitHub: https://github.com/paramjeetn/RagBench⁠
• Website & Documentation: https://ragbench-web.vercel.app⁠

⁠📄 License

See the LICENSE⁠ file for licensing information.