Memgraph GenAI Stack

Docker Compose + FastAPI + LangChain + Ollama + Memgraph

This demo is a template for building GenAI applications with Memgraph.

App architecture

Dataset

In order for schema to be generated, a dataset needs to be loaded in Memgraph before the application is started. The Game of Thrones dataset is preloaded in this demo. To preload your own dataset, update the dataset.cypherl file. The first three lines in the CYPHERL file are there, so the data doesn't reload on each run. (STORAGE MODE IN_MEMORY_ANALYTICAL; DROP GRAPH; STORAGE MODE IN_MEMORY_TRANSACTIONAL;)

Start the app

To utilize GPT-4 model, create the .env file in the root directory and set your OpenAI API key:

OPENAI_API_KEY=<YOUR_OPENAI_API_KEY>

Run the app:

docker compose up

This demo offers querying Memgraph with the natural language via LangChain utilizing two different models:

• GPT-4
• Llama 3

Ask Memgraph with GPT-4

Ask Memgraph with GPT-4 model:

curl -X POST "http://localhost:8000/ask/gpt" -H "Content-Type: application/json" -d '{"question": "How many seasons there are?"}'

Here is the response:

{"question":"How many seasons there are?","response":"There are 8 seasons."}% 

Ask Memgraph with Llama 3

Ask Memgraph with Llama3 model:

curl -X POST "http://localhost:8000/ask/llama" -H "Content-Type: application/json" -d '{"question": "How many seasons there are?"}' 

Here is the response:

{"question":"How many seasons there are?","response":"There are 8 seasons."}%