title	emoji	colorFrom	colorTo	sdk	sdk_version	app_file	pinned
Hybrid RAG	📚	blue	green	streamlit	1.49.1	app.py	false

RAG (Retrieval-Augmented Generation)

A lightweight hybrid RAG prototype combining retrieval and language model generation to answer queries over documents or knowledge sources.

Accesible in Hugging Face space - https://huggingface.co/spaces/polojuan/hybridrag

---

🚀 Features

• Hybrid retrieval (semantic + keyword-based) over document corpus
• Connects retrieval output into an LLM prompt for answer generation
• Simple, modular architecture for easy experimentation
• Docker support for reproducible environments
• Python-based, minimal dependencies

---

Architecture & Workflow

1. Document ingestion & preprocessing

   • Load documents (e.g. PDF, text)
   • Chunk / split into manageable passages

2. Embedding & indexing

   • Generate embeddings for chunks using e.g. sentence-transformer
   • Store embeddings in a vector index

3. Retrieval (hybrid)

   • Semantic (vector) search
   • Keyword / sparse search (optional)
   • Fuse or rerank results

4. Prompt construction & generation

   • Construct the prompt combining query + retrieved context
   • Send to a language model (e.g. via OpenAI API or local LLM)
   • Return answer

5. (Optional) Postprocessing & filtering

   • Clean up output, optionally verify or validate

---

🛠️ Setup & Installation

Prerequisites

• Python 3.8+
• Docker (if you choose to use container)

Local setup

git clone https://github.com/palscruz23/rag.git
cd rag

# Install dependencies
pip install -r requirements.txt

# Run streamlit app
streamlit run app.py

---

📁 Repository Structure

.
├── app.py              # Main entry / API or UI driver
├── Dockerfile          # Docker configuration
├── requirements.txt    # Python dependencies
├── utils/              # Utility modules & helpers
└── README.md           # This documentation

You may also have submodules under utils (e.g. embedding, retrieval, prompt, generation).

---

📜 License

This repository is licensed under the MIT License. See the LICENSE file for more details.