IS Theory RAG Prototype

A minimal Retrieval-Augmented Generation (RAG) prototype that indexes Information Systems theories into a local vector database and answers natural-language queries with the most semantically relevant theories.

Everything runs locally in a single Jupyter notebook — no API keys, no external services. It uses:

• Weaviate (embedded) — an in-process vector database; no server to run.
• FastEmbed with the BAAI/bge-small-en-v1.5 model — computes embeddings locally.

This is a learning prototype built while following the DeepLearning.AI × Astronomer course Orchestrating Workflows for GenAI Applications (see Acknowledgments).

Prerequisites

• Python ≥ 3.12
• The repo ships with a checked-in venv/. If you prefer a fresh environment:

  python3 -m venv venv
  source venv/bin/activate
  pip install -r notebooks/requirements.txt

Usage

Run from the notebooks/ directory (paths in the notebook are relative to it):

source venv/bin/activate
cd notebooks
jupyter lab        # then open prototype.ipynb and run the cells top-to-bottom

The notebook, in order:

1. Starts an embedded Weaviate instance (data persists to notebooks/tmp/weaviate/).
2. Creates a Theories collection.
3. Parses every theory file in notebooks/include/data/ into {name, description, authors, seminal_articles}.
4. Embeds each theory (name + description) and loads the vectors into Weaviate.
5. Runs a semantic query and prints the top matching theories.

Change the query_str variable in the query cell to search for different theories.

Note: Embedded Weaviate binds ports 8079/50050. If a previous run left a process behind, a new run fails with a "ports already listening" error — clear it with pkill -f weaviate.

Data format

Each notebooks/include/data/*.txt file is one theory in Markdown. The # heading is the theory name; the ## sections hold the content (any field may be N/A):

# <Theory name>
## Concise description of theory
<description or N/A>
## Originating author(s)
<authors or N/A>
## Seminal articles
<articles or N/A>

License

The code in this repository is released under the MIT License.

The MIT license covers only the code. The theory descriptions under notebooks/include/data/ are third-party content (see Acknowledgments) and remain subject to their original terms.

Acknowledgments

• Course & structure — This prototype follows the RAG walkthrough from the Orchestrating Workflows for GenAI Applications short course by DeepLearning.AI in partnership with Astronomer, taught by Kenten Danas and Tamara Fingerlin. It is an independent reimplementation for learning purposes (the original example indexes book descriptions; this version indexes IS theories).
• Theory data — The theory descriptions were scraped from the Information Systems Theories wiki (is.theorizeit.org). Credit for that content belongs to the wiki and its contributors.