Home

Awesome

Knowledge Graph Builder App

Creating knowledge graphs from unstructured data

LLM Graph Builder

Python FastAPI React

Overview

This application is designed to turn Unstructured data (pdfs,docs,txt,youtube video,web pages,etc.) into a knowledge graph stored in Neo4j. It utilizes the power of Large language models (OpenAI,Gemini,etc.) to extract nodes, relationships and their properties from the text and create a structured knowledge graph using Langchain framework.

Upload your files from local machine, GCS or S3 bucket or from web sources, choose your LLM model and generate knowledge graph.

Key Features

Getting started

:warning: You will need to have a Neo4j Database V5.15 or later with APOC installed to use this Knowledge Graph Builder. You can use any Neo4j Aura database (including the free database) If you are using Neo4j Desktop, you will not be able to use the docker-compose but will have to follow the separate deployment of backend and frontend section. :warning:

Deployment

Local deployment

Running through docker-compose

By default only OpenAI and Diffbot are enabled since Gemini requires extra GCP configurations. According to enviornment we are configuring the models which is indicated by VITE_LLM_MODELS_PROD variable we can configure model based on our need. EX:

VITE_LLM_MODELS_PROD="openai_gpt_4o,openai_gpt_4o_mini,diffbot,gemini_1.5_flash"

OPENAI_API_KEY="your-openai-key"


You can then run Docker Compose to build and start all components:
```bash
docker-compose up --build

Additional configs

By default, the input sources will be: Local files, Youtube, Wikipedia ,AWS S3 and Webpages. As this default config is applied:

VITE_REACT_APP_SOURCES="local,youtube,wiki,s3,web"

If however you want the Google GCS integration, add gcs and your Google client ID:

VITE_REACT_APP_SOURCES="local,youtube,wiki,s3,gcs,web"
VITE_GOOGLE_CLIENT_ID="xxxx"

You can of course combine all (local, youtube, wikipedia, s3 and gcs) or remove any you don't want/need.

Chat Modes

By default,all of the chat modes will be available: vector, graph_vector, graph, fulltext, graph_vector_fulltext , entity_vector and global_vector.

If none of the mode is mentioned in the chat modes variable all modes will be available:

VITE_CHAT_MODES=""

If however you want to specify the only vector mode or only graph mode you can do that by specifying the mode in the env:

VITE_CHAT_MODES="vector,graph"
VITE_CHAT_MODES="vector,graph"

Running Backend and Frontend separately (dev environment)

Alternatively, you can run the backend and frontend separately:

  1. Create the frontend/.env file by copy/pasting the frontend/example.env.
  2. Change values as needed
  3. cd frontend
    yarn
    yarn run dev
    
  1. Create the backend/.env file by copy/pasting the backend/example.env. To streamline the initial setup and testing of the application, you can preconfigure user credentials directly within the .env file. This bypasses the login dialog and allows you to immediately connect with a predefined user.
    • NEO4J_URI:
    • NEO4J_USERNAME:
    • NEO4J_PASSWORD:
    • NEO4J_DATABASE:
  2. Change values as needed
  3. cd backend
    python -m venv envName
    source envName/bin/activate 
    pip install -r requirements.txt
    uvicorn score:app --reload
    

Deploy in Cloud

To deploy the app and packages on Google Cloud Platform, run the following command on google cloud run:

# Frontend deploy 
gcloud run deploy dev-frontend 
source location current directory > Frontend
region : 32 [us-central 1]
Allow unauthenticated request : Yes
# Backend deploy 
gcloud run deploy --set-env-vars "OPENAI_API_KEY = " --set-env-vars "DIFFBOT_API_KEY = " --set-env-vars "NEO4J_URI = " --set-env-vars "NEO4J_PASSWORD = " --set-env-vars "NEO4J_USERNAME = "
source location current directory > Backend
region : 32 [us-central 1]
Allow unauthenticated request : Yes

ENV

Env Variable NameMandatory/OptionalDefault ValueDescription
EMBEDDING_MODELOptionalall-MiniLM-L6-v2Model for generating the text embedding (all-MiniLM-L6-v2 , openai , vertexai)
IS_EMBEDDINGOptionaltrueFlag to enable text embedding
KNN_MIN_SCOREOptional0.94Minimum score for KNN algorithm
GEMINI_ENABLEDOptionalFalseFlag to enable Gemini
GCP_LOG_METRICS_ENABLEDOptionalFalseFlag to enable Google Cloud logs
NUMBER_OF_CHUNKS_TO_COMBINEOptional5Number of chunks to combine when processing embeddings
UPDATE_GRAPH_CHUNKS_PROCESSEDOptional20Number of chunks processed before updating progress
NEO4J_URIOptionalneo4j://database:7687URI for Neo4j database
NEO4J_USERNAMEOptionalneo4jUsername for Neo4j database
NEO4J_PASSWORDOptionalpasswordPassword for Neo4j database
LANGCHAIN_API_KEYOptionalAPI key for Langchain
LANGCHAIN_PROJECTOptionalProject for Langchain
LANGCHAIN_TRACING_V2OptionaltrueFlag to enable Langchain tracing
LANGCHAIN_ENDPOINTOptionalhttps://api.smith.langchain.comEndpoint for Langchain API
VITE_BACKEND_API_URLOptionalhttp://localhost:8000URL for backend API
VITE_BLOOM_URLOptionalhttps://workspace-preview.neo4j.io/workspace/explore?connectURL={CONNECT_URL}&search=Show+me+a+graph&featureGenAISuggestions=true&featureGenAISuggestionsInternal=trueURL for Bloom visualization
VITE_REACT_APP_SOURCESMandatorylocal,youtube,wiki,s3List of input sources that will be available
VITE_CHAT_MODESMandatoryvector,graph+vector,graph,hybridChat modes available for Q&A
VITE_ENVMandatoryDEV or PRODEnvironment variable for the app
VITE_TIME_PER_PAGEOptional50Time per page for processing
VITE_CHUNK_SIZEOptional5242880Size of each chunk of file for upload
VITE_GOOGLE_CLIENT_IDOptionalClient ID for Google authentication
VITE_LLM_MODELS_PRODOptionalopenai_gpt_4o,openai_gpt_4o_mini,diffbot,gemini_1.5_flashTo Distinguish models based on the Enviornment PROD or DEV
VITE_LLM_MODELSOptional'diffbot,openai_gpt_3.5,openai_gpt_4o,openai_gpt_4o_mini,gemini_1.5_pro,gemini_1.5_flash,azure_ai_gpt_35,azure_ai_gpt_4o,ollama_llama3,groq_llama3_70b,anthropic_claude_3_5_sonnet'Supported Models For the application
GCS_FILE_CACHEOptionalFalseIf set to True, will save the files to process into GCS. If set to False, will save the files locally
ENTITY_EMBEDDINGOptionalFalseIf set to True, It will add embeddings for each entity in database
LLM_MODEL_CONFIG_ollama_<model_name>OptionalSet ollama config as - model_name,model_local_url for local deployments
RAGAS_EMBEDDING_MODELOptionalopenaiembedding model used by ragas evaluation framework

LLMs Supported

  1. OpenAI
  2. Gemini
  3. Azure OpenAI(dev)
  4. Anthropic(dev)
  5. Fireworks(dev)
  6. Groq(dev)
  7. Amazon Bedrock(dev)
  8. Ollama(dev)
  9. Diffbot
  10. Other OpenAI compabtile baseurl models(dev)

For local llms (Ollama)

  1. Pull the docker imgage of ollama
docker pull ollama/ollama
  1. Run the ollama docker image
docker run -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama
  1. Execute any llm model ex🦙3
docker exec -it ollama ollama run llama3
  1. Configure env variable in docker compose or backend environment.
LLM_MODEL_CONFIG_ollama_<model_name>
#example
LLM_MODEL_CONFIG_ollama_llama3=${LLM_MODEL_CONFIG_ollama_llama3-llama3,
http://host.docker.internal:11434}
  1. Configure the backend API url
VITE_BACKEND_API_URL=${VITE_BACKEND_API_URL-backendurl}
  1. Open the application in browser and select the ollama model for the extraction.
  2. Enjoy Graph Building.

Usage

  1. Connect to Neo4j Aura Instance which can be both AURA DS or AURA DB by passing URI and password through Backend env, fill using login dialog or drag and drop the Neo4j credentials file.
  2. To differntiate we have added different icons. For AURA DB we have a database icon and for AURA DS we have scientific molecule icon right under Neo4j Connection details label.
  3. Choose your source from a list of Unstructured sources to create graph.
  4. Change the LLM (if required) from drop down, which will be used to generate graph.
  5. Optionally, define schema(nodes and relationship labels) in entity graph extraction settings.
  6. Either select multiple files to 'Generate Graph' or all the files in 'New' status will be processed for graph creation.
  7. Have a look at the graph for individual files using 'View' in grid or select one or more files and 'Preview Graph'
  8. Ask questions related to the processed/completed sources to chat-bot, Also get detailed information about your answers generated by LLM.

Links

LLM Knowledge Graph Builder Application

Neo4j Workspace

Reference

Demo of application

Contact

For any inquiries or support, feel free to raise Github Issue

Happy Graph Building!