Basic RAG

Retrieval‑Augmented Generation demo with LangChain, FAISS and OpenAI

Features

• Downloads Wikipedia articles on a topic of your choice via WikipediaLoader.
• Splits text into semantically coherent chunks with SemanticChunker.
• Creates an in‑memory FAISS vector store with OpenAI embeddings.
• Retrieves the most relevant chunks for a user question.
• Generates an answer with GPT-4o using a prompt pulled from LangChain Hub.
• Everything runs in a single script: naive_rag_wikipedia.py.

Quick start

# 1. Clone the repo
git clone https://github.com/felipeortizh/basic-RAG.git
cd basic-RAG

# 2. (Optional) create & activate a virtual environment
python -m venv .venv && source .venv/bin/activate      # macOS / Linux
# or on Windows
python -m venv .venv && .\.venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

Environment variables

Set your OpenAI API key before running:

export OPENAI_API_KEY="sk-..."            # macOS / Linux
setx OPENAI_API_KEY "sk-..."              # Windows

Run the demo

python naive_rag_wikipedia.py

Sample output:

Starting document loading...
Documents loaded successfully: 2 documents
Creating text splits...
Created 34 text splits
Creating vector store...
Vector store created successfully
...
Result: The main areas of Artificial Intelligence are...

How it works

1. Ingest – WikipediaLoader fetches up to n articles for the query defined in the script (default: "Artificial Intelligence").
2. Chunk – SemanticChunker breaks each article into ~500‑token, semantically coherent chunks.
3. Index – Each chunk is embedded with OpenAIEmbeddings and stored in a FAISS index.
4. Retrieve – The top‑k chunks relevant to the user question are retrieved.
5. Generate – Chunks are injected into a RAG prompt template from LangChain Hub and sent to the LLM for the final answer.

Wikipedia → Chunk → FAISS → Retriever → Prompt → LLM

Customisation

Change you want	Where to edit
Topic to search on Wikipedia	query argument of WikipediaLoader
Number of articles to load	load_max_docs
Embedding model	OpenAIEmbeddings()
Vector store implementation	replace FAISS with any LangChain store
LLM model	ChatOpenAI(model_name="gpt-3.5-turbo")

Repository structure

basic-RAG/
├── naive_rag_wikipedia.py   # end‑to‑end RAG pipeline
├── requirements.txt         # Python dependencies
└── README.md                # You are here

Roadmap

• Parse CLI arguments for topic, doc count & model selection
• Add persistence to the FAISS index
• Stream token-by-token LLM output
• Unit tests

Contributing

Contributions are welcome! Please open an issue to discuss your ideas or submit a pull request.

License

Released under the MIT License.