scribe_rag is a naive RAG application that was developed as my final bachelor's project at university.
scribe_rag consists of two parts:
scribe- Python web serverscribe_ui- TypeScript web interface
scribe_rag allows for full customization of the RAG pipeline:
- Language models from various providers
- Embedding models from various providers, including open-source, locally hosted options
- Simple and semantic chunking strategy customization
- System prompt and language model parameters
- Embedding model and parameters
- Vector collection and base chat customizations
- Various document file formats supported
- Open and clear context that is ingested into a model
Here are some screenshots of app usage.
System prompt configuration:
Chat model configuration:
Embedding model configuration:
Semantic chunking configuration:
Vector collection configuration and document ingestion (the request was: How to format formulas?):
Base chat creation:
Finally, communication in a chat with ingested documents:
To clone the project, run:
git clone https://github.com/panalexeu/scribe_rag.git
The server is fully written in Python, hence it depends on the package manager. As the package manager, Poetry was chosen, so to run the server, Poetry must be installed.
Run:
cd ./scribe
Then:
poetry install
To install all dependent packages. After that, you must activate the Python shell with:
poetry shell
Great, the only thing left is to simply run:
./start.sh
To start the server and Chroma DB.
To run the server, envs must be provided in the .env file:
- SCRIBE_DB - Configure SQLite DB type: dev - in-memory, prod - in-file.
Makefile stores commands to test and run the server.
The UI is fully written in JS/TypeScript and depends on the npm package manager. Hence npm is a must to run the UI.
Run:
cd ./scribe_ui
Then:
npm install
To install all dependent packages. After that, you can start the UI with:
npm run dev