Advanced RAG Cookbooks

Three practical retrieval projects spanning vector similarity, local Ollama RAG, and a document-chat API with a Streamlit client.

Projects · Quick starts · Screenshot · Limitations

What is it?

This repository is a collection of independent Retrieval-Augmented Generation experiments, not one installable package. Each directory has its own dependencies, runtime assumptions, and data path.

The projects cover a notebook-driven book recommender, a local Llama 3 assistant backed by PostgreSQL/pgvector, and a FastAPI + Chroma document-chat service with a Streamlit UI.

Project catalog

Project	Interface	Retrieval / model path	Evidence in the repository
book-recommender-vector-similarity/	Jupyter notebooks	Tagged descriptions and vector similarity over the included book data	Two notebooks, dataset files, and a captured Python environment
rag-llama3-ollama-webapp/	Streamlit	Ollama Llama 3 + Ollama embeddings + PostgreSQL/pgvector	App, assistant configuration, and setup notes
rag-streamlit-fastapi/	FastAPI and Streamlit	Local Hugging Face embeddings + Chroma + Groq chat model	API, client, sample PDF, persisted Chroma data, API PDF, and screenshots

Real project screenshot

The image below is the checked-in Streamlit client for rag-streamlit-fastapi after processing a document and starting a chat.

Quick starts

Book recommender

cd book-recommender-vector-similarity
python -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
jupyter lab

Open data-explore.ipynb first, then vector-search.ipynb. The requirements file is a broad captured environment rather than a minimal dependency set.

Local Llama 3 + pgvector

Prerequisites: Python, Docker, and Ollama with the llama3 model available.

cd rag-llama3-ollama-webapp
python -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
ollama pull llama3

Start the database expected by assistant.py:

docker run -d \
  --name pgvector \
  -e POSTGRES_DB=ai \
  -e POSTGRES_USER=ai \
  -e POSTGRES_PASSWORD=ai \
  -p 5532:5432 \
  phidata/pgvector:16

Then run:

streamlit run app.py

FastAPI + Chroma document chat

cd rag-streamlit-fastapi
python -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt

Create .env with a Groq key:

GROQ_API_KEY=your_key_here

Run the API and UI in separate terminals:

python main.py

streamlit run streamlit.py

The API documentation is served at http://localhost:8000/docs. The Dockerfile runs the API only; the Streamlit client is a separate process.

Status and limitations

• The projects are independent experiments with no root dependency lock or shared runner.
• rag-streamlit-fastapi accepts server-local file paths, keeps chat sessions in memory, enables wildcard CORS, and includes a persisted Chroma database; review these choices before using untrusted data or exposing it publicly.
• Its Streamlit client deletes the temporary upload immediately after the API call and loses chat state on reload.
• The local Ollama project uses the older phi package API and a hard-coded PostgreSQL URL.
• The book recommender environment is large and not guaranteed to reproduce across platforms.
• No repository-wide license, CI configuration, or automated test suite is present.